20231088/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Docs] Convert rST to MyST (Markdown) (#11145)

Browse Source 
Signed-off-by: Rafael Vasquez <rafvasq21@gmail.com>
This commit is contained in:
Rafael Vasquez 2024-12-23 17:35:38 -05:00 committed by GitHub
parent 94d545a1a1
commit 32aa2059ad
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
167 changed files with 7863 additions and 8131 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored
View File

@ -81,6 +81,8 @@ instance/
docs/_build/
docs/source/getting_started/examples/*.rst
!**/*.template.rst
docs/source/getting_started/examples/*.md
!**/*.template.md
# PyBuilder
.pybuilder/

2
Dockerfile
View File

@ -2,7 +2,7 @@
# to run the OpenAI compatible server.
# Please update any changes made here to
# docs/source/dev/dockerfile/dockerfile.rst and
# docs/source/dev/dockerfile/dockerfile.md and
# docs/source/assets/dev/dockerfile-stages-dependency.png
ARG CUDA_VERSION=12.4.1

2
docs/requirements-docs.txt
View File

@ -1,7 +1,7 @@
sphinx==6.2.1
sphinx-book-theme==1.0.1
sphinx-copybutton==0.5.2
myst-parser==2.0.0
myst-parser==3.0.1
sphinx-argparse==0.4.0
msgspec
cloudpickle

102
docs/source/automatic_prefix_caching/apc.md Normal file
View File

@ -0,0 +1,102 @@
(apc)=
# Introduction
## What is Automatic Prefix Caching
Automatic Prefix Caching (APC in short) caches the KV cache of existing queries, so that a new query can directly reuse the KV cache if it shares the same prefix with one of the existing queries, allowing the new query to skip the computation of the shared part.
```{note}
Technical details on how vLLM implements APC are in the next page.
```
## Enabling APC in vLLM
Set `enable_prefix_caching=True` in vLLM engine to enable APC. Here is an example:
```python
import time
from vllm import LLM, SamplingParams
# A prompt containing a large markdown table. The table is randomly generated by GPT-4.
LONG_PROMPT = "You are a helpful assistant in recognizes the content of tables in markdown format. Here is a table as follows.\n# Table\n" + """
| ID | Name | Age | Occupation | Country | Email | Phone Number | Address |
|-----|---------------|-----|---------------|---------------|------------------------|----------------|------------------------------|
| 1 | John Doe | 29 | Engineer | USA | john.doe@example.com | 555-1234 | 123 Elm St, Springfield, IL |
| 2 | Jane Smith | 34 | Doctor | Canada | jane.smith@example.com | 555-5678 | 456 Oak St, Toronto, ON |
| 3 | Alice Johnson | 27 | Teacher | UK | alice.j@example.com | 555-8765 | 789 Pine St, London, UK |
| 4 | Bob Brown | 45 | Artist | Australia | bob.b@example.com | 555-4321 | 321 Maple St, Sydney, NSW |
| 5 | Carol White | 31 | Scientist | New Zealand | carol.w@example.com | 555-6789 | 654 Birch St, Wellington, NZ |
| 6 | Dave Green | 28 | Lawyer | Ireland | dave.g@example.com | 555-3456 | 987 Cedar St, Dublin, IE |
| 7 | Emma Black | 40 | Musician | USA | emma.b@example.com | 555-1111 | 246 Ash St, New York, NY |
| 8 | Frank Blue | 37 | Chef | Canada | frank.b@example.com | 555-2222 | 135 Spruce St, Vancouver, BC |
| 9 | Grace Yellow | 50 | Engineer | UK | grace.y@example.com | 555-3333 | 864 Fir St, Manchester, UK |
| 10 | Henry Violet | 32 | Artist | Australia | henry.v@example.com | 555-4444 | 753 Willow St, Melbourne, VIC|
| 11 | Irene Orange | 26 | Scientist | New Zealand | irene.o@example.com | 555-5555 | 912 Poplar St, Auckland, NZ |
| 12 | Jack Indigo | 38 | Teacher | Ireland | jack.i@example.com | 555-6666 | 159 Elm St, Cork, IE |
| 13 | Karen Red | 41 | Lawyer | USA | karen.r@example.com | 555-7777 | 357 Cedar St, Boston, MA |
| 14 | Leo Brown | 30 | Chef | Canada | leo.b@example.com | 555-8888 | 246 Oak St, Calgary, AB |
| 15 | Mia Green | 33 | Musician | UK | mia.g@example.com | 555-9999 | 975 Pine St, Edinburgh, UK |
| 16 | Noah Yellow | 29 | Doctor | Australia | noah.y@example.com | 555-0000 | 864 Birch St, Brisbane, QLD |
| 17 | Olivia Blue | 35 | Engineer | New Zealand | olivia.b@example.com | 555-1212 | 753 Maple St, Hamilton, NZ |
| 18 | Peter Black | 42 | Artist | Ireland | peter.b@example.com | 555-3434 | 912 Fir St, Limerick, IE |
| 19 | Quinn White | 28 | Scientist | USA | quinn.w@example.com | 555-5656 | 159 Willow St, Seattle, WA |
| 20 | Rachel Red | 31 | Teacher | Canada | rachel.r@example.com | 555-7878 | 357 Poplar St, Ottawa, ON |
| 21 | Steve Green | 44 | Lawyer | UK | steve.g@example.com | 555-9090 | 753 Elm St, Birmingham, UK |
| 22 | Tina Blue | 36 | Musician | Australia | tina.b@example.com | 555-1213 | 864 Cedar St, Perth, WA |
| 23 | Umar Black | 39 | Chef | New Zealand | umar.b@example.com | 555-3435 | 975 Spruce St, Christchurch, NZ|
| 24 | Victor Yellow | 43 | Engineer | Ireland | victor.y@example.com | 555-5657 | 246 Willow St, Galway, IE |
| 25 | Wendy Orange | 27 | Artist | USA | wendy.o@example.com | 555-7879 | 135 Elm St, Denver, CO |
| 26 | Xavier Green | 34 | Scientist | Canada | xavier.g@example.com | 555-9091 | 357 Oak St, Montreal, QC |
| 27 | Yara Red | 41 | Teacher | UK | yara.r@example.com | 555-1214 | 975 Pine St, Leeds, UK |
| 28 | Zack Blue | 30 | Lawyer | Australia | zack.b@example.com | 555-3436 | 135 Birch St, Adelaide, SA |
| 29 | Amy White | 33 | Musician | New Zealand | amy.w@example.com | 555-5658 | 159 Maple St, Wellington, NZ |
| 30 | Ben Black | 38 | Chef | Ireland | ben.b@example.com | 555-7870 | 246 Fir St, Waterford, IE |
"""
def get_generation_time(llm, sampling_params, prompts):
# time the generation
start_time = time.time()
output = llm.generate(prompts, sampling_params=sampling_params)
end_time = time.time()
# print the output and generation time
print(f"Output: {output[0].outputs[0].text}")
print(f"Generation time: {end_time - start_time} seconds.")
# set enable_prefix_caching=True to enable APC
llm = LLM(
model='lmsys/longchat-13b-16k',
enable_prefix_caching=True
)
sampling_params = SamplingParams(temperature=0, max_tokens=100)
# Querying the age of John Doe
get_generation_time(
llm,
sampling_params,
LONG_PROMPT + "Question: what is the age of John Doe? Your answer: The age of John Doe is ",
)
# Querying the age of Zack Blue
# This query will be faster since vllm avoids computing the KV cache of LONG_PROMPT again.
get_generation_time(
llm,
sampling_params,
LONG_PROMPT + "Question: what is the age of Zack Blue? Your answer: The age of Zack Blue is ",
)
```
## Example workloads
We describe two example workloads, where APC can provide huge performance benefit:
- Long document query, where the user repeatedly queries the same long document (e.g. software manual or annual report) with different queries. In this case, instead of processing the long document again and again, APC allows vLLM to process this long document *only once*, and all future requests can avoid recomputing this long document by reusing its KV cache. This allows vLLM to serve future requests with much higher throughput and much lower latency.
- Multi-round conversation, where the user may chat with the application multiple times in the same chatting session. In this case, instead of processing the whole chatting history again and again, APC allows vLLM to reuse the processing results of the chat history across all future rounds of conversation, allowing vLLM to serve future requests with much higher throughput and much lower latency.
## Limits
APC in general does not reduce the performance of vLLM. With that being said, APC only reduces the time of processing the queries (the prefilling phase) and does not reduce the time of generating new tokens (the decoding phase). So APC does not bring performance gain when vLLM spends most of the time generating answers to the queries (e.g. when the length of the answer is long), or new queries do not share the same prefix with any of existing queries (so that the computation cannot be reused).

110
docs/source/automatic_prefix_caching/apc.rst
View File

@ -1,110 +0,0 @@
.. _apc:
Introduction
============
What is Automatic Prefix Caching
--------------------------------
Automatic Prefix Caching (APC in short) caches the KV cache of existing queries, so that a new query can directly reuse the KV cache if it shares the same prefix with one of the existing queries, allowing the new query to skip the computation of the shared part.
.. note::
Technical details on how vLLM implements APC are in the next page.
Enabling APC in vLLM
--------------------
Set ``enable_prefix_caching=True`` in vLLM engine to enable APC. Here is an example:
.. code-block:: python
import time
from vllm import LLM, SamplingParams
# A prompt containing a large markdown table. The table is randomly generated by GPT-4.
LONG_PROMPT = "You are a helpful assistant in recognizes the content of tables in markdown format. Here is a table as follows.\n# Table\n" + """
| ID | Name | Age | Occupation | Country | Email | Phone Number | Address |
|-----|---------------|-----|---------------|---------------|------------------------|----------------|------------------------------|
| 1 | John Doe | 29 | Engineer | USA | john.doe@example.com | 555-1234 | 123 Elm St, Springfield, IL |
| 2 | Jane Smith | 34 | Doctor | Canada | jane.smith@example.com | 555-5678 | 456 Oak St, Toronto, ON |
| 3 | Alice Johnson | 27 | Teacher | UK | alice.j@example.com | 555-8765 | 789 Pine St, London, UK |
| 4 | Bob Brown | 45 | Artist | Australia | bob.b@example.com | 555-4321 | 321 Maple St, Sydney, NSW |
| 5 | Carol White | 31 | Scientist | New Zealand | carol.w@example.com | 555-6789 | 654 Birch St, Wellington, NZ |
| 6 | Dave Green | 28 | Lawyer | Ireland | dave.g@example.com | 555-3456 | 987 Cedar St, Dublin, IE |
| 7 | Emma Black | 40 | Musician | USA | emma.b@example.com | 555-1111 | 246 Ash St, New York, NY |
| 8 | Frank Blue | 37 | Chef | Canada | frank.b@example.com | 555-2222 | 135 Spruce St, Vancouver, BC |
| 9 | Grace Yellow | 50 | Engineer | UK | grace.y@example.com | 555-3333 | 864 Fir St, Manchester, UK |
| 10 | Henry Violet | 32 | Artist | Australia | henry.v@example.com | 555-4444 | 753 Willow St, Melbourne, VIC|
| 11 | Irene Orange | 26 | Scientist | New Zealand | irene.o@example.com | 555-5555 | 912 Poplar St, Auckland, NZ |
| 12 | Jack Indigo | 38 | Teacher | Ireland | jack.i@example.com | 555-6666 | 159 Elm St, Cork, IE |
| 13 | Karen Red | 41 | Lawyer | USA | karen.r@example.com | 555-7777 | 357 Cedar St, Boston, MA |
| 14 | Leo Brown | 30 | Chef | Canada | leo.b@example.com | 555-8888 | 246 Oak St, Calgary, AB |
| 15 | Mia Green | 33 | Musician | UK | mia.g@example.com | 555-9999 | 975 Pine St, Edinburgh, UK |
| 16 | Noah Yellow | 29 | Doctor | Australia | noah.y@example.com | 555-0000 | 864 Birch St, Brisbane, QLD |
| 17 | Olivia Blue | 35 | Engineer | New Zealand | olivia.b@example.com | 555-1212 | 753 Maple St, Hamilton, NZ |
| 18 | Peter Black | 42 | Artist | Ireland | peter.b@example.com | 555-3434 | 912 Fir St, Limerick, IE |
| 19 | Quinn White | 28 | Scientist | USA | quinn.w@example.com | 555-5656 | 159 Willow St, Seattle, WA |
| 20 | Rachel Red | 31 | Teacher | Canada | rachel.r@example.com | 555-7878 | 357 Poplar St, Ottawa, ON |
| 21 | Steve Green | 44 | Lawyer | UK | steve.g@example.com | 555-9090 | 753 Elm St, Birmingham, UK |
| 22 | Tina Blue | 36 | Musician | Australia | tina.b@example.com | 555-1213 | 864 Cedar St, Perth, WA |
| 23 | Umar Black | 39 | Chef | New Zealand | umar.b@example.com | 555-3435 | 975 Spruce St, Christchurch, NZ|
| 24 | Victor Yellow | 43 | Engineer | Ireland | victor.y@example.com | 555-5657 | 246 Willow St, Galway, IE |
| 25 | Wendy Orange | 27 | Artist | USA | wendy.o@example.com | 555-7879 | 135 Elm St, Denver, CO |
| 26 | Xavier Green | 34 | Scientist | Canada | xavier.g@example.com | 555-9091 | 357 Oak St, Montreal, QC |
| 27 | Yara Red | 41 | Teacher | UK | yara.r@example.com | 555-1214 | 975 Pine St, Leeds, UK |
| 28 | Zack Blue | 30 | Lawyer | Australia | zack.b@example.com | 555-3436 | 135 Birch St, Adelaide, SA |
| 29 | Amy White | 33 | Musician | New Zealand | amy.w@example.com | 555-5658 | 159 Maple St, Wellington, NZ |
| 30 | Ben Black | 38 | Chef | Ireland | ben.b@example.com | 555-7870 | 246 Fir St, Waterford, IE |
"""
def get_generation_time(llm, sampling_params, prompts):
# time the generation
start_time = time.time()
output = llm.generate(prompts, sampling_params=sampling_params)
end_time = time.time()
# print the output and generation time
print(f"Output: {output[0].outputs[0].text}")
print(f"Generation time: {end_time - start_time} seconds.")
# set enable_prefix_caching=True to enable APC
llm = LLM(
model='lmsys/longchat-13b-16k',
enable_prefix_caching=True
)
sampling_params = SamplingParams(temperature=0, max_tokens=100)
# Querying the age of John Doe
get_generation_time(
llm,
sampling_params,
LONG_PROMPT + "Question: what is the age of John Doe? Your answer: The age of John Doe is ",
)
# Querying the age of Zack Blue
# This query will be faster since vllm avoids computing the KV cache of LONG_PROMPT again.
get_generation_time(
llm,
sampling_params,
LONG_PROMPT + "Question: what is the age of Zack Blue? Your answer: The age of Zack Blue is ",
)
Example workloads
-----------------
We describe two example workloads, where APC can provide huge performance benefit:
- Long document query, where the user repeatedly queries the same long document (e.g. software manual or annual report) with different queries. In this case, instead of processing the long document again and again, APC allows vLLM to process this long document *only once*, and all future requests can avoid recomputing this long document by reusing its KV cache. This allows vLLM to serve future requests with much higher throughput and much lower latency.
- Multi-round conversation, where the user may chat with the application multiple times in the same chatting session. In this case, instead of processing the whole chatting history again and again, APC allows vLLM to reuse the processing results of the chat history across all future rounds of conversation, allowing vLLM to serve future requests with much higher throughput and much lower latency.
Limits
------
APC in general does not reduce the performance of vLLM. With that being said, APC only reduces the time of processing the queries (the prefilling phase) and does not reduce the time of generating new tokens (the decoding phase). So APC does not bring performance gain when vLLM spends most of the time generating answers to the queries (e.g. when the length of the answer is long), or new queries do not share the same prefix with any of existing queries (so that the computation cannot be reused).

15
docs/source/community/meetups.md Normal file
View File

@ -0,0 +1,15 @@
(meetups)=
# vLLM Meetups
We host regular meetups in San Francisco Bay Area every 2 months. We will share the project updates from the vLLM team and have guest speakers from the industry to share their experience and insights. Please find the materials of our previous meetups below:
- [The seventh vLLM meetup](https://lu.ma/h0qvrajz), with Snowflake, November 14th 2024. [[Slides]](https://docs.google.com/presentation/d/1e3CxQBV3JsfGp30SwyvS3eM_tW-ghOhJ9PAJGK6KR54/edit?usp=sharing)
- [The sixth vLLM meetup](https://lu.ma/87q3nvnh), with NVIDIA, September 9th 2024. [[Slides]](https://docs.google.com/presentation/d/1wrLGwytQfaOTd5wCGSPNhoaW3nq0E-9wqyP7ny93xRs/edit?usp=sharing)
- [The fifth vLLM meetup](https://lu.ma/lp0gyjqr), with AWS, July 24th 2024. [[Slides]](https://docs.google.com/presentation/d/1RgUD8aCfcHocghoP3zmXzck9vX3RCI9yfUAB2Bbcl4Y/edit?usp=sharing)
- [The fourth vLLM meetup](https://lu.ma/agivllm), with Cloudflare and BentoML, June 11th 2024. [[Slides]](https://docs.google.com/presentation/d/1iJ8o7V2bQEi0BFEljLTwc5G1S10_Rhv3beed5oB0NJ4/edit?usp=sharing)
- [The third vLLM meetup](https://robloxandvllmmeetup2024.splashthat.com/), with Roblox, April 2nd 2024. [[Slides]](https://docs.google.com/presentation/d/1A--47JAK4BJ39t954HyTkvtfwn0fkqtsL8NGFuslReM/edit?usp=sharing)
- [The second vLLM meetup](https://lu.ma/ygxbpzhl), with IBM Research, January 31st 2024. [[Slides]](https://docs.google.com/presentation/d/12mI2sKABnUw5RBWXDYY-HtHth4iMSNcEoQ10jDQbxgA/edit?usp=sharing) [[Video (vLLM Update)]](https://youtu.be/Y0C-DUvEnZQ) [[Video (IBM Research & torch.compile)]](https://youtu.be/m0dMtFLI-dg)
- [The first vLLM meetup](https://lu.ma/first-vllm-meetup), with a16z, October 5th 2023. [[Slides]](https://docs.google.com/presentation/d/1QL-XPFXiFpDBh86DbEegFXBXFXjix4v032GhShbKf3s/edit?usp=sharing)
We are always looking for speakers and sponsors at San Francisco Bay Area and potentially other locations. If you are interested in speaking or sponsoring, please contact us at [vllm-questions@lists.berkeley.edu](mailto:vllm-questions@lists.berkeley.edu).

16
docs/source/community/meetups.rst
View File

@ -1,16 +0,0 @@
.. _meetups:
vLLM Meetups
============
We host regular meetups in San Francisco Bay Area every 2 months. We will share the project updates from the vLLM team and have guest speakers from the industry to share their experience and insights. Please find the materials of our previous meetups below:
- `The seventh vLLM meetup <https://lu.ma/h0qvrajz>`__, with Snowflake, November 14th 2024. `[Slides] <https://docs.google.com/presentation/d/1e3CxQBV3JsfGp30SwyvS3eM_tW-ghOhJ9PAJGK6KR54/edit?usp=sharing>`__
- `The sixth vLLM meetup <https://lu.ma/87q3nvnh>`__, with NVIDIA, September 9th 2024. `[Slides] <https://docs.google.com/presentation/d/1wrLGwytQfaOTd5wCGSPNhoaW3nq0E-9wqyP7ny93xRs/edit?usp=sharing>`__
- `The fifth vLLM meetup <https://lu.ma/lp0gyjqr>`__, with AWS, July 24th 2024. `[Slides] <https://docs.google.com/presentation/d/1RgUD8aCfcHocghoP3zmXzck9vX3RCI9yfUAB2Bbcl4Y/edit?usp=sharing>`__
- `The fourth vLLM meetup <https://lu.ma/agivllm>`__, with Cloudflare and BentoML, June 11th 2024. `[Slides] <https://docs.google.com/presentation/d/1iJ8o7V2bQEi0BFEljLTwc5G1S10_Rhv3beed5oB0NJ4/edit?usp=sharing>`__
- `The third vLLM meetup <https://robloxandvllmmeetup2024.splashthat.com/>`__, with Roblox, April 2nd 2024. `[Slides] <https://docs.google.com/presentation/d/1A--47JAK4BJ39t954HyTkvtfwn0fkqtsL8NGFuslReM/edit?usp=sharing>`__
- `The second vLLM meetup <https://lu.ma/ygxbpzhl>`__, with IBM Research, January 31st 2024. `[Slides] <https://docs.google.com/presentation/d/12mI2sKABnUw5RBWXDYY-HtHth4iMSNcEoQ10jDQbxgA/edit?usp=sharing>`__ `[Video (vLLM Update)] <https://youtu.be/Y0C-DUvEnZQ>`__ `[Video (IBM Research & torch.compile)] <https://youtu.be/m0dMtFLI-dg>`__
- `The first vLLM meetup <https://lu.ma/first-vllm-meetup>`__, with a16z, October 5th 2023. `[Slides] <https://docs.google.com/presentation/d/1QL-XPFXiFpDBh86DbEegFXBXFXjix4v032GhShbKf3s/edit?usp=sharing>`__
We are always looking for speakers and sponsors at San Francisco Bay Area and potentially other locations. If you are interested in speaking or sponsoring, please contact us at `vllm-questions@lists.berkeley.edu <mailto:vllm-questions@lists.berkeley.edu>`__.

2
docs/source/conf.py
View File

@ -51,7 +51,7 @@ templates_path = ['_templates']
# List of patterns, relative to source directory, that match files and
# directories to ignore when looking for source files.
# This pattern also affects html_static_path and html_extra_path.
exclude_patterns: List[str] = ["**/*.template.rst"]
exclude_patterns: List[str] = ["**/*.template.md"]
# Exclude the prompt "$" when copying code
copybutton_prompt_text = r"\$ "

50
docs/source/contributing/dockerfile/dockerfile.md Normal file
View File

@ -0,0 +1,50 @@
# Dockerfile
See [here](https://github.com/vllm-project/vllm/blob/main/Dockerfile) for the main Dockerfile to construct
the image for running an OpenAI compatible server with vLLM. More information about deploying with Docker can be found [here](https://docs.vllm.ai/en/stable/serving/deploying_with_docker.html).
Below is a visual representation of the multi-stage Dockerfile. The build graph contains the following nodes:
- All build stages
- The default build target (highlighted in grey)
- External images (with dashed borders)
The edges of the build graph represent:
- FROM ... dependencies (with a solid line and a full arrow head)
- COPY --from=... dependencies (with a dashed line and an empty arrow head)
- RUN --mount=(.\*)from=... dependencies (with a dotted line and an empty diamond arrow head)
> ```{figure} ../../assets/dev/dockerfile-stages-dependency.png
> :align: center
> :alt: query
> :width: 100%
> ```
>
> Made using: <https://github.com/patrickhoefler/dockerfilegraph>
>
> Commands to regenerate the build graph (make sure to run it **from the \`root\` directory of the vLLM repository** where the dockerfile is present):
>
> ```bash
> dockerfilegraph -o png --legend --dpi 200 --max-label-length 50 --filename Dockerfile
> ```
>
> or in case you want to run it directly with the docker image:
>
> ```bash
> docker run \
> --rm \
> --user "$(id -u):$(id -g)" \
> --workdir /workspace \
> --volume "$(pwd)":/workspace \
> ghcr.io/patrickhoefler/dockerfilegraph:alpine \
> --output png \
> --dpi 200 \
> --max-label-length 50 \
> --filename Dockerfile \
> --legend
> ```
>
> (To run it for a different file, you can pass in a different argument to the flag `--filename`.)

50
docs/source/contributing/dockerfile/dockerfile.rst
View File

@ -1,50 +0,0 @@
Dockerfile
====================
See `here <https://github.com/vllm-project/vllm/blob/main/Dockerfile>`__ for the main Dockerfile to construct
the image for running an OpenAI compatible server with vLLM. More information about deploying with Docker can be found `here <https://docs.vllm.ai/en/stable/serving/deploying_with_docker.html>`__.
Below is a visual representation of the multi-stage Dockerfile. The build graph contains the following nodes:
- All build stages
- The default build target (highlighted in grey)
- External images (with dashed borders)
The edges of the build graph represent:
- FROM ... dependencies (with a solid line and a full arrow head)
- COPY --from=... dependencies (with a dashed line and an empty arrow head)
- RUN --mount=(.*)from=... dependencies (with a dotted line and an empty diamond arrow head)
.. figure:: ../../assets/dev/dockerfile-stages-dependency.png
:alt: query
:width: 100%
:align: center
Made using: https://github.com/patrickhoefler/dockerfilegraph
Commands to regenerate the build graph (make sure to run it **from the `root` directory of the vLLM repository** where the dockerfile is present):
.. code:: bash
dockerfilegraph -o png --legend --dpi 200 --max-label-length 50 --filename Dockerfile
or in case you want to run it directly with the docker image:
.. code:: bash
docker run \
--rm \
--user "$(id -u):$(id -g)" \
--workdir /workspace \
--volume "$(pwd)":/workspace \
ghcr.io/patrickhoefler/dockerfilegraph:alpine \
--output png \
--dpi 200 \
--max-label-length 50 \
--filename Dockerfile \
--legend
(To run it for a different file, you can pass in a different argument to the flag `--filename`.)

137
docs/source/contributing/overview.rst → docs/source/contributing/overview.md
View File

@ -1,5 +1,4 @@
Contributing to vLLM
=====================
# Contributing to vLLM
Thank you for your interest in contributing to vLLM! Our community is open to everyone and welcomes all kinds of contributions, no matter how small or large. There are several ways you can contribute to the project:
@ -12,132 +11,121 @@ We also believe in the power of community support; thus, answering queries, offe
Finally, one of the most impactful ways to support us is by raising awareness about vLLM. Talk about it in your blog posts and highlight how it's driving your incredible projects. Express your support on social media if you're using vLLM, or simply offer your appreciation by starring our repository!
License
-------
## License
See `LICENSE <https://github.com/vllm-project/vllm/tree/main/LICENSE>`_.
See [LICENSE](https://github.com/vllm-project/vllm/tree/main/LICENSE).
Developing
----------
## Developing
Depending on the kind of development you'd like to do (e.g. Python, CUDA), you can choose to build vLLM with or without compilation. Check out the `building from source <https://docs.vllm.ai/en/latest/getting_started/installation.html#build-from-source>`_ documentation for details.
Depending on the kind of development you'd like to do (e.g. Python, CUDA), you can choose to build vLLM with or without compilation. Check out the [building from source](https://docs.vllm.ai/en/latest/getting_started/installation.html#build-from-source) documentation for details.
Testing
-------
## Testing
.. code-block:: bash
```bash
pip install -r requirements-dev.txt
pip install -r requirements-dev.txt
# linting and formatting
bash format.sh
# Static type checking
mypy
# Unit tests
pytest tests/
```
# linting and formatting
bash format.sh
# Static type checking
mypy
# Unit tests
pytest tests/
```{note}
Currently, the repository does not pass the `mypy` tests.
```
.. note:: Currently, the repository does not pass the ``mypy`` tests.
# Contribution Guidelines
Contribution Guidelines
=======================
## Issues
Issues
------
If you encounter a bug or have a feature request, please [search existing issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue) first to see if it has already been reported. If not, please [file a new issue](https://github.com/vllm-project/vllm/issues/new/choose), providing as much relevant information as possible.
If you encounter a bug or have a feature request, please `search existing issues <https://github.com/vllm-project/vllm/issues?q=is%3Aissue>`_ first to see if it has already been reported. If not, please `file a new issue <https://github.com/vllm-project/vllm/issues/new/choose>`_, providing as much relevant information as possible.
```{important}
If you discover a security vulnerability, please follow the instructions [here](https://github.com/vllm-project/vllm/tree/main/SECURITY.md#reporting-a-vulnerability).
```
.. important::
If you discover a security vulnerability, please follow the instructions `here <https://github.com/vllm-project/vllm/tree/main/SECURITY.md#reporting-a-vulnerability>`_.
Pull Requests & Code Reviews
----------------------------
## Pull Requests & Code Reviews
Thank you for your contribution to vLLM! Before submitting the pull request,
please ensure the PR meets the following criteria. This helps vLLM maintain the
code quality and improve the efficiency of the review process.
DCO and Signed-off-by
^^^^^^^^^^^^^^^^^^^^^
### DCO and Signed-off-by
When contributing changes to this project, you must agree to the `DCO <https://github.com/vllm-project/vllm/tree/main/DCO>`_.
Commits must include a ``Signed-off-by:`` header which certifies agreement with
the terms of the `DCO <https://github.com/vllm-project/vllm/tree/main/DCO>`_.
When contributing changes to this project, you must agree to the [DCO](https://github.com/vllm-project/vllm/tree/main/DCO).
Commits must include a `Signed-off-by:` header which certifies agreement with
the terms of the [DCO](https://github.com/vllm-project/vllm/tree/main/DCO).
Using ``-s`` with ``git commit`` will automatically add this header.
Using `-s` with `git commit` will automatically add this header.
PR Title and Classification
^^^^^^^^^^^^^^^^^^^^^^^^^^^
### PR Title and Classification
Only specific types of PRs will be reviewed. The PR title is prefixed
appropriately to indicate the type of change. Please use one of the following:
- ``[Bugfix]`` for bug fixes.
- ``[CI/Build]`` for build or continuous integration improvements.
- ``[Doc]`` for documentation fixes and improvements.
- ``[Model]`` for adding a new model or improving an existing model. Model name
- `[Bugfix]` for bug fixes.
- `[CI/Build]` for build or continuous integration improvements.
- `[Doc]` for documentation fixes and improvements.
- `[Model]` for adding a new model or improving an existing model. Model name
should appear in the title.
- ``[Frontend]`` For changes on the vLLM frontend (e.g., OpenAI API server,
``LLM`` class, etc.)
- ``[Kernel]`` for changes affecting CUDA kernels or other compute kernels.
- ``[Core]`` for changes in the core vLLM logic (e.g., ``LLMEngine``,
``AsyncLLMEngine``, ``Scheduler``, etc.)
- ``[Hardware][Vendor]`` for hardware-specific changes. Vendor name should
appear in the prefix (e.g., ``[Hardware][AMD]``).
- ``[Misc]`` for PRs that do not fit the above categories. Please use this
- `[Frontend]` For changes on the vLLM frontend (e.g., OpenAI API server,
`LLM` class, etc.)
- `[Kernel]` for changes affecting CUDA kernels or other compute kernels.
- `[Core]` for changes in the core vLLM logic (e.g., `LLMEngine`,
`AsyncLLMEngine`, `Scheduler`, etc.)
- `[Hardware][Vendor]` for hardware-specific changes. Vendor name should
appear in the prefix (e.g., `[Hardware][AMD]`).
- `[Misc]` for PRs that do not fit the above categories. Please use this
sparingly.
.. note::
If the PR spans more than one category, please include all relevant prefixes.
```{note}
If the PR spans more than one category, please include all relevant prefixes.
```
Code Quality
^^^^^^^^^^^^
### Code Quality
The PR needs to meet the following code quality standards:
- We adhere to `Google Python style guide
<https://google.github.io/styleguide/pyguide.html>`_ and `Google C++ style guide
<https://google.github.io/styleguide/cppguide.html>`_.
- Pass all linter checks. Please use `format.sh
<https://github.com/vllm-project/vllm/blob/main/format.sh>`_ to format your
- We adhere to [Google Python style guide](https://google.github.io/styleguide/pyguide.html) and [Google C++ style guide](https://google.github.io/styleguide/cppguide.html).
- Pass all linter checks. Please use [format.sh](https://github.com/vllm-project/vllm/blob/main/format.sh) to format your
code.
- The code needs to be well-documented to ensure future contributors can easily
understand the code.
- Include sufficient tests to ensure the project stays correct and robust. This
includes both unit tests and integration tests.
- Please add documentation to ``docs/source/`` if the PR modifies the
- Please add documentation to `docs/source/` if the PR modifies the
user-facing behaviors of vLLM. It helps vLLM users understand and utilize the
new features or changes.
Adding or Changing Kernels
^^^^^^^^^^^^^^^^^^^^^^^^^^
### Adding or Changing Kernels
Each custom kernel needs a schema and one or more implementations to be registered with PyTorch.
- Make sure custom ops are registered following PyTorch guidelines:
`Custom C++ and CUDA Operators <https://pytorch.org/tutorials/advanced/cpp_custom_ops.html#cpp-custom-ops-tutorial>`_
and `The Custom Operators Manual <https://docs.google.com/document/d/1_W62p8WJOQQUzPsJYa7s701JXt0qf2OfLub2sbkHOaU>`_.
- Custom operations that return ``Tensors`` require meta-functions.
[Custom C++ and CUDA Operators](https://pytorch.org/tutorials/advanced/cpp_custom_ops.html#cpp-custom-ops-tutorial)
and [The Custom Operators Manual](https://docs.google.com/document/d/1_W62p8WJOQQUzPsJYa7s701JXt0qf2OfLub2sbkHOaU).
- Custom operations that return `Tensors` require meta-functions.
Meta-functions should be implemented and registered in Python so that dynamic
dims can be handled automatically. See above documents for a description of
meta-functions.
- Use `torch.library.opcheck() <https://pytorch.org/docs/stable/library.html#torch.library.opcheck>`_
- Use [torch.library.opcheck()](https://pytorch.org/docs/stable/library.html#torch.library.opcheck)
to test the function registration and meta-function for any registered ops.
See ``tests/kernels`` for examples.
See `tests/kernels` for examples.
- When changing the C++ signature of an existing op, the schema must be updated
to reflect the changes.
- If a new custom type is needed, see the following document:
`Custom Class Support in PT2 <https://docs.google.com/document/d/18fBMPuOJ0fY5ZQ6YyrHUppw9FA332CpNtgB6SOIgyuA>`_.
[Custom Class Support in PT2](https://docs.google.com/document/d/18fBMPuOJ0fY5ZQ6YyrHUppw9FA332CpNtgB6SOIgyuA).
Notes for Large Changes
^^^^^^^^^^^^^^^^^^^^^^^
### Notes for Large Changes
Please keep the changes as concise as possible. For major architectural changes
(>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue
(RFC) discussing the technical design and justification. Otherwise, we will tag
it with ``rfc-required`` and might not go through the PR.
it with `rfc-required` and might not go through the PR.
What to Expect for the Reviews
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
### What to Expect for the Reviews
The goal of the vLLM team is to be a *transparent reviewing machine*. We would
like to make the review process transparent and efficient and make sure no
@ -150,15 +138,14 @@ review process:
- After the PR is assigned, the reviewer will provide status updates every 2-3
days. If the PR is not reviewed within 7 days, please feel free to ping the
reviewer or the vLLM team.
- After the review, the reviewer will put an ``action-required`` label on the PR
- After the review, the reviewer will put an `action-required` label on the PR
if there are changes required. The contributor should address the comments and
ping the reviewer to re-review the PR.
- Please respond to all comments within a reasonable time frame. If a comment
isn't clear or you disagree with a suggestion, feel free to ask for
clarification or discuss the suggestion.
Thank You
---------
## Thank You
Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM.
All of your contributions help make vLLM a great tool and community for everyone!

41
docs/source/contributing/profiling/profiling_index.md Normal file
View File

@ -0,0 +1,41 @@
# Profiling vLLM
We support tracing vLLM workers using the `torch.profiler` module. You can enable tracing by setting the `VLLM_TORCH_PROFILER_DIR` environment variable to the directory where you want to save the traces: `VLLM_TORCH_PROFILER_DIR=/mnt/traces/`
The OpenAI server also needs to be started with the `VLLM_TORCH_PROFILER_DIR` environment variable set.
When using `benchmarks/benchmark_serving.py`, you can enable profiling by passing the `--profile` flag.
```{warning}
Only enable profiling in a development environment.
```
Traces can be visualized using <https://ui.perfetto.dev/>.
```{tip}
Only send a few requests through vLLM when profiling, as the traces can get quite large. Also, no need to untar the traces, they can be viewed directly.
```
```{tip}
To stop the profiler - it flushes out all the profile trace files to the directory. This takes time, for example for about 100 requests worth of data for a llama 70b, it takes about 10 minutes to flush out on a H100.
Set the env variable VLLM_RPC_TIMEOUT to a big number before you start the server. Say something like 30 minutes.
`export VLLM_RPC_TIMEOUT=1800000`
```
## Example commands and usage:
### Offline Inference:
Refer to [examples/offline_inference_with_profiler.py](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_with_profiler.py) for an example.
### OpenAI Server:
```bash
VLLM_TORCH_PROFILER_DIR=./vllm_profile python -m vllm.entrypoints.openai.api_server --model meta-llama/Meta-Llama-3-70B
```
benchmark_serving.py:
```bash
python benchmarks/benchmark_serving.py --backend vllm --model meta-llama/Meta-Llama-3-70B --dataset-name sharegpt --dataset-path sharegpt.json --profile --num-prompts 2
```

48
docs/source/contributing/profiling/profiling_index.rst
View File

@ -1,48 +0,0 @@
==============
Profiling vLLM
==============
We support tracing vLLM workers using the ``torch.profiler`` module. You can enable tracing by setting the ``VLLM_TORCH_PROFILER_DIR`` environment variable to the directory where you want to save the traces: ``VLLM_TORCH_PROFILER_DIR=/mnt/traces/``
The OpenAI server also needs to be started with the ``VLLM_TORCH_PROFILER_DIR`` environment variable set.
When using ``benchmarks/benchmark_serving.py``, you can enable profiling by passing the ``--profile`` flag.
.. warning::
Only enable profiling in a development environment.
Traces can be visualized using https://ui.perfetto.dev/.
.. tip::
Only send a few requests through vLLM when profiling, as the traces can get quite large. Also, no need to untar the traces, they can be viewed directly.
.. tip::
To stop the profiler - it flushes out all the profile trace files to the directory. This takes time, for example for about 100 requests worth of data for a llama 70b, it takes about 10 minutes to flush out on a H100.
Set the env variable VLLM_RPC_TIMEOUT to a big number before you start the server. Say something like 30 minutes.
``export VLLM_RPC_TIMEOUT=1800000``
Example commands and usage:
===========================
Offline Inference:
------------------
Refer to `examples/offline_inference_with_profiler.py <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_with_profiler.py>`_ for an example.
OpenAI Server:
--------------
.. code-block:: bash
VLLM_TORCH_PROFILER_DIR=./vllm_profile python -m vllm.entrypoints.openai.api_server --model meta-llama/Meta-Llama-3-70B
benchmark_serving.py:
.. code-block:: bash
python benchmarks/benchmark_serving.py --backend vllm --model meta-llama/Meta-Llama-3-70B --dataset-name sharegpt --dataset-path sharegpt.json --profile --num-prompts 2

236
docs/source/design/arch_overview.rst → docs/source/design/arch_overview.md
View File

@ -1,25 +1,24 @@
.. _arch_overview:
(arch-overview)=
Architecture Overview
======================
# Architecture Overview
This document provides an overview of the vLLM architecture.
.. contents:: Table of Contents
:local:
:depth: 2
```{contents} Table of Contents
:depth: 2
:local: true
```
Entrypoints
-----------
## Entrypoints
vLLM provides a number of entrypoints for interacting with the system. The
following diagram shows the relationship between them.
.. image:: /assets/design/arch_overview/entrypoints.excalidraw.png
:alt: Entrypoints Diagram
```{image} /assets/design/arch_overview/entrypoints.excalidraw.png
:alt: Entrypoints Diagram
```
LLM Class
^^^^^^^^^
### LLM Class
The LLM class provides the primary Python interface for doing offline inference,
which is interacting with a model without using a separate model inference
@ -27,75 +26,70 @@ server.
Here is a sample of `LLM` class usage:
.. code-block:: python
```python
from vllm import LLM, SamplingParams
from vllm import LLM, SamplingParams
# Define a list of input prompts
prompts = [
"Hello, my name is",
"The capital of France is",
"The largest ocean is",
]
# Define a list of input prompts
prompts = [
"Hello, my name is",
"The capital of France is",
"The largest ocean is",
]
# Define sampling parameters
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Define sampling parameters
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Initialize the LLM engine with the OPT-125M model
llm = LLM(model="facebook/opt-125m")
# Initialize the LLM engine with the OPT-125M model
llm = LLM(model="facebook/opt-125m")
# Generate outputs for the input prompts
outputs = llm.generate(prompts, sampling_params)
# Generate outputs for the input prompts
outputs = llm.generate(prompts, sampling_params)
# Print the generated outputs
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
# Print the generated outputs
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
More API details can be found in the :doc:`Offline Inference
More API details can be found in the {doc}`Offline Inference
</dev/offline_inference/offline_index>` section of the API docs.
The code for the `LLM` class can be found in `vllm/entrypoints/llm.py
<https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/llm.py>`_.
The code for the `LLM` class can be found in [vllm/entrypoints/llm.py](https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/llm.py).
OpenAI-compatible API server
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
### OpenAI-compatible API server
The second primary interface to vLLM is via its OpenAI-compatible API server.
This server can be started using the `vllm serve` command.
.. code-block:: bash
```bash
vllm serve <model>
```
vllm serve <model>
The code for the `vllm` CLI can be found in `vllm/scripts.py
<https://github.com/vllm-project/vllm/blob/main/vllm/scripts.py>`_.
The code for the `vllm` CLI can be found in [vllm/scripts.py](https://github.com/vllm-project/vllm/blob/main/vllm/scripts.py).
Sometimes you may see the API server entrypoint used directly instead of via the
`vllm` CLI command. For example:
.. code-block:: bash
```bash
python -m vllm.entrypoints.openai.api_server --model <model>
```
python -m vllm.entrypoints.openai.api_server --model <model>
That code can be found in [vllm/entrypoints/openai/api_server.py](https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/openai/api_server.py).
That code can be found in `vllm/entrypoints/openai/api_server.py
<https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/openai/api_server.py>`_.
More details on the API server can be found in the :doc:`OpenAI Compatible
More details on the API server can be found in the {doc}`OpenAI Compatible
Server </serving/openai_compatible_server>` document.
LLM Engine
----------
## LLM Engine
The `LLMEngine` and `AsyncLLMEngine` classes are central to the functioning of
the vLLM system, handling model inference and asynchronous request processing.
.. image:: /assets/design/arch_overview/llm_engine.excalidraw.png
:alt: LLMEngine Diagram
```{image} /assets/design/arch_overview/llm_engine.excalidraw.png
:alt: LLMEngine Diagram
```
LLMEngine
^^^^^^^^^
### LLMEngine
The `LLMEngine` class is the core component of the vLLM engine. It is
responsible for receiving requests from clients and generating outputs from the
@ -105,21 +99,15 @@ processing.
- **Input Processing**: Handles tokenization of input text using the specified
tokenizer.
- **Scheduling**: Chooses which requests are processed in each step.
- **Model Execution**: Manages the execution of the language model, including
distributed execution across multiple GPUs.
- **Output Processing**: Processes the outputs generated by the model, decoding the
token IDs from a language model into human-readable text.
The code for `LLMEngine` can be found in `vllm/engine/llm_engine.py`_.
The code for `LLMEngine` can be found in [vllm/engine/llm_engine.py].
.. _vllm/engine/llm_engine.py: https://github.com/vllm-project/vllm/tree/main/vllm/engine/llm_engine.py
AsyncLLMEngine
^^^^^^^^^^^^^^
### AsyncLLMEngine
The `AsyncLLMEngine` class is an asynchronous wrapper for the `LLMEngine` class.
It uses `asyncio` to create a background loop that continuously processes
@ -128,54 +116,46 @@ can handle multiple concurrent requests and stream outputs to clients.
The OpenAI-compatible API server uses the `AsyncLLMEngine`. There is also a demo
API server that serves as a simpler example in
`vllm/entrypoints/api_server.py`_.
[vllm/entrypoints/api_server.py].
.. _vllm/entrypoints/api_server.py: https://github.com/vllm-project/vllm/tree/main/vllm/entrypoints/api_server.py
The code for `AsyncLLMEngine` can be found in [vllm/engine/async_llm_engine.py].
The code for `AsyncLLMEngine` can be found in `vllm/engine/async_llm_engine.py`_.
.. _vllm/engine/async_llm_engine.py: https://github.com/vllm-project/vllm/tree/main/vllm/engine/async_llm_engine.py
Worker
------
## Worker
A worker is a process that runs the model inference. vLLM follows the common
practice of using one process to control one accelerator device, such as GPUs.
For example, if we use tensor parallelism of size 2 and pipeline parallelism of
size 2, we will have 4 workers in total. Workers are identified by their
``rank`` and ``local_rank``. ``rank`` is used for global orchestration, while
``local_rank`` is mainly used for assigning the accelerator device and accessing
`rank` and `local_rank`. `rank` is used for global orchestration, while
`local_rank` is mainly used for assigning the accelerator device and accessing
local resources such as the file system and shared memory.
Model Runner
------------
## Model Runner
Every worker has one model runner object, responsible for loading and running
the model. Much of the model execution logic resides here, such as preparing
input tensors and capturing cudagraphs.
Model
-----
## Model
Every model runner object has one model object, which is the actual
``torch.nn.Module`` instance. See :ref:`huggingface_integration` for how various
`torch.nn.Module` instance. See [huggingface_integration](#huggingface-integration) for how various
configurations affect the class we ultimately get.
Class Hierarchy
---------------
## Class Hierarchy
The following figure shows the class hierarchy of vLLM:
.. figure:: /assets/design/hierarchy.png
:alt: query
:width: 100%
:align: center
> ```{figure} /assets/design/hierarchy.png
> :align: center
> :alt: query
> :width: 100%
> ```
There are several important design choices behind this class hierarchy:
1. **Extensibility**: All classes in the hierarchy accept a configuration object
containing all the necessary information. The `VllmConfig
<https://github.com/vllm-project/vllm/blob/d1c6799b8870e513bf4f2305cbf6cda9fc3d773b/vllm/config.py#L2036>`__
1\. **Extensibility**: All classes in the hierarchy accept a configuration object
containing all the necessary information. The [VllmConfig](https://github.com/vllm-project/vllm/blob/d1c6799b8870e513bf4f2305cbf6cda9fc3d773b/vllm/config.py#L2036)
class is the main configuration object that is passed around. The class
hierarchy is quite deep, and every class needs to read the configuration it is
interested in. By encapsulating all configurations in one object, we can easily
@ -188,7 +168,7 @@ the `VllmConfig` class, and the model runner can access it directly. We don't
need to change the constructor of the engine, worker, or model class to pass the
new configuration option.
2. **Uniformity**: The model runner needs a unified interface to create and
2\. **Uniformity**: The model runner needs a unified interface to create and
initialize the model. vLLM supports more than 50 types of popular open-source
models. Each model has its own initialization logic. If the constructor
signature varies with models, the model runner does not know how to call the
@ -200,46 +180,46 @@ of a vision model and a language model. By making the constructor uniform, we
can easily create a vision model and a language model and compose them into a
vision-language model.
.. note::
````{note}
To support this change, all vLLM models' signatures have been updated to:
To support this change, all vLLM models' signatures have been updated to:
```python
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
```
.. code-block:: python
To avoid accidentally passing incorrect arguments, the constructor is now keyword-only. This ensures that the constructor will raise an error if old configurations are passed. vLLM developers have already made this change for all models within vLLM. For out-of-tree registered models, developers need to update their models, for example by adding shim code to adapt the old constructor signature to the new one:
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
```python
class MyOldModel(nn.Module):
def __init__(
self,
config,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
lora_config: Optional[LoRAConfig] = None,
prefix: str = "",
) -> None:
...
To avoid accidentally passing incorrect arguments, the constructor is now keyword-only. This ensures that the constructor will raise an error if old configurations are passed. vLLM developers have already made this change for all models within vLLM. For out-of-tree registered models, developers need to update their models, for example by adding shim code to adapt the old constructor signature to the new one:
from vllm.config import VllmConfig
class MyNewModel(MyOldModel):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
config = vllm_config.model_config.hf_config
cache_config = vllm_config.cache_config
quant_config = vllm_config.quant_config
lora_config = vllm_config.lora_config
super().__init__(config, cache_config, quant_config, lora_config, prefix)
.. code-block:: python
if __version__ >= "0.6.4":
MyModel = MyNewModel
else:
MyModel = MyOldModel
```
class MyOldModel(nn.Module):
def __init__(
self,
config,
cache_config: Optional[CacheConfig] = None,
quant_config: Optional[QuantizationConfig] = None,
lora_config: Optional[LoRAConfig] = None,
prefix: str = "",
) -> None:
...
This way, the model can work with both old and new versions of vLLM.
````
from vllm.config import VllmConfig
class MyNewModel(MyOldModel):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
config = vllm_config.model_config.hf_config
cache_config = vllm_config.cache_config
quant_config = vllm_config.quant_config
lora_config = vllm_config.lora_config
super().__init__(config, cache_config, quant_config, lora_config, prefix)
if __version__ >= "0.6.4":
MyModel = MyNewModel
else:
MyModel = MyOldModel
This way, the model can work with both old and new versions of vLLM.
3. **Sharding and Quantization at Initialization**: Certain features require
3\. **Sharding and Quantization at Initialization**: Certain features require
changing the model weights. For example, tensor parallelism needs to shard the
model weights, and quantization needs to quantize the model weights. There are
two possible ways to implement this feature. One way is to change the model
@ -252,23 +232,27 @@ initialized, we need to load the full 810GB weights to every GPU and then shard
the weights, leading to a huge memory overhead. Instead, if we shard the weights
during the model initialization, every layer will only create a shard of the
weights it needs, leading to a much smaller memory overhead. The same idea
applies to quantization. Note that we also add an additional argument ``prefix``
applies to quantization. Note that we also add an additional argument `prefix`
to the model's constructor so that the model can initialize itself differently
based on the prefix. This is useful for non-uniform quantization, where
different parts of the model are quantized differently. The ``prefix`` is
usually an empty string for the top-level model and a string like ``"vision"``
or ``"language"`` for the sub-models. In general, it matches the name of the
different parts of the model are quantized differently. The `prefix` is
usually an empty string for the top-level model and a string like `"vision"`
or `"language"` for the sub-models. In general, it matches the name of the
module's state dict in the checkpoint file.
One disadvantage of this design is that it is hard to write unit tests for
individual components in vLLM because every component needs to be initialized by
a complete config object. We solve this problem by providing a default
initialization function that creates a default config object with all fields set
to ``None``. If the component we want to test only cares about a few fields in
to `None`. If the component we want to test only cares about a few fields in
the config object, we can create a default config object and set the fields we
care about. This way, we can test the component in isolation. Note that many
tests in vLLM are end-to-end tests that test the whole system, so this is not a
big problem.
In summary, the complete config object ``VllmConfig`` can be treated as an
In summary, the complete config object `VllmConfig` can be treated as an
engine-level global state that is shared among all vLLM classes.
[vllm/engine/async_llm_engine.py]: https://github.com/vllm-project/vllm/tree/main/vllm/engine/async_llm_engine.py
[vllm/engine/llm_engine.py]: https://github.com/vllm-project/vllm/tree/main/vllm/engine/llm_engine.py
[vllm/entrypoints/api_server.py]: https://github.com/vllm-project/vllm/tree/main/vllm/entrypoints/api_server.py

36
docs/source/design/huggingface_integration.md Normal file
View File

@ -0,0 +1,36 @@
(huggingface-integration)=
# Integration with HuggingFace
This document describes how vLLM integrates with HuggingFace libraries. We will explain step by step what happens under the hood when we run `vllm serve`.
Let's say we want to serve the popular QWen model by running `vllm serve Qwen/Qwen2-7B`.
1. The `model` argument is `Qwen/Qwen2-7B`. vLLM determines whether this model exists by checking for the corresponding config file `config.json`. See this [code snippet](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L162-L182) for the implementation. Within this process:
- If the `model` argument corresponds to an existing local path, vLLM will load the config file directly from this path.
- If the `model` argument is a HuggingFace model ID consisting of a username and model name, vLLM will first try to use the config file from the HuggingFace local cache, using the `model` argument as the model name and the `--revision` argument as the revision. See [their website](https://huggingface.co/docs/huggingface_hub/en/package_reference/environment_variables#hfhome) for more information on how the HuggingFace cache works.
- If the `model` argument is a HuggingFace model ID but it is not found in the cache, vLLM will download the config file from the HuggingFace model hub. Refer to [this function](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L91) for the implementation. The input arguments include the `model` argument as the model name, the `--revision` argument as the revision, and the environment variable `HF_TOKEN` as the token to access the model hub. In our case, vLLM will download the [config.json](https://huggingface.co/Qwen/Qwen2-7B/blob/main/config.json) file.
2. After confirming the existence of the model, vLLM loads its config file and converts it into a dictionary. See this [code snippet](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L185-L186) for the implementation.
3. Next, vLLM [inspects](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L189) the `model_type` field in the config dictionary to [generate](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#190-L216) the config object to use. There are some `model_type` values that vLLM directly supports; see [here](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L48) for the list. If the `model_type` is not in the list, vLLM will use [AutoConfig.from_pretrained](https://huggingface.co/docs/transformers/en/model_doc/auto#transformers.AutoConfig.from_pretrained) to load the config class, with `model`, `--revision`, and `--trust_remote_code` as the arguments. Please note that:
- HuggingFace also has its own logic to determine the config class to use. It will again use the `model_type` field to search for the class name in the transformers library; see [here](https://github.com/huggingface/transformers/tree/main/src/transformers/models) for the list of supported models. If the `model_type` is not found, HuggingFace will use the `auto_map` field from the config JSON file to determine the class name. Specifically, it is the `AutoConfig` field under `auto_map`. See [DeepSeek](https://huggingface.co/deepseek-ai/DeepSeek-V2.5/blob/main/config.json) for an example.
- The `AutoConfig` field under `auto_map` points to a module path in the model's repository. To create the config class, HuggingFace will import the module and use the `from_pretrained` method to load the config class. This can generally cause arbitrary code execution, so it is only executed when `--trust_remote_code` is enabled.
4. Subsequently, vLLM applies some historical patches to the config object. These are mostly related to RoPE configuration; see [here](https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/transformers_utils/config.py#L244) for the implementation.
5. Finally, vLLM can reach the model class we want to initialize. vLLM uses the `architectures` field in the config object to determine the model class to initialize, as it maintains the mapping from architecture name to model class in [its registry](https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/model_executor/models/registry.py#L80). If the architecture name is not found in the registry, it means this model architecture is not supported by vLLM. For `Qwen/Qwen2-7B`, the `architectures` field is `["Qwen2ForCausalLM"]`, which corresponds to the `Qwen2ForCausalLM` class in [vLLM's code](https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/model_executor/models/qwen2.py#L364). This class will initialize itself depending on various configs.
Beyond that, there are two more things vLLM depends on HuggingFace for.
1. **Tokenizer**: vLLM uses the tokenizer from HuggingFace to tokenize the input text. The tokenizer is loaded using [AutoTokenizer.from_pretrained](https://huggingface.co/docs/transformers/en/model_doc/auto#transformers.AutoTokenizer.from_pretrained) with the `model` argument as the model name and the `--revision` argument as the revision. It is also possible to use a tokenizer from another model by specifying the `--tokenizer` argument in the `vllm serve` command. Other relevant arguments are `--tokenizer-revision` and `--tokenizer-mode`. Please check HuggingFace's documentation for the meaning of these arguments. This part of the logic can be found in the [get_tokenizer](https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/transformers_utils/tokenizer.py#L87) function. After obtaining the tokenizer, notably, vLLM will cache some expensive attributes of the tokenizer in [get_cached_tokenizer](https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/transformers_utils/tokenizer.py#L24).
2. **Model weight**: vLLM downloads the model weight from the HuggingFace model hub using the `model` argument as the model name and the `--revision` argument as the revision. vLLM provides the argument `--load-format` to control what files to download from the model hub. By default, it will try to load the weights in the safetensors format and fall back to the PyTorch bin format if the safetensors format is not available. We can also pass `--load-format dummy` to skip downloading the weights.
- It is recommended to use the safetensors format, as it is efficient for loading in distributed inference and also safe from arbitrary code execution. See the [documentation](https://huggingface.co/docs/safetensors/en/index) for more information on the safetensors format. This part of the logic can be found [here](https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/model_executor/model_loader/loader.py#L385). Please note that:
This completes the integration between vLLM and HuggingFace.
In summary, vLLM reads the config file `config.json`, tokenizer, and model weight from the HuggingFace model hub or a local directory. It uses the config class from either vLLM, HuggingFace transformers, or loads the config class from the model's repository.

40
docs/source/design/huggingface_integration.rst
View File

@ -1,40 +0,0 @@
.. _huggingface_integration:
Integration with HuggingFace
===================================
This document describes how vLLM integrates with HuggingFace libraries. We will explain step by step what happens under the hood when we run ``vllm serve``.
Let's say we want to serve the popular QWen model by running ``vllm serve Qwen/Qwen2-7B``.
1. The ``model`` argument is ``Qwen/Qwen2-7B``. vLLM determines whether this model exists by checking for the corresponding config file ``config.json``. See this `code snippet <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L162-L182>`__ for the implementation. Within this process:
- If the ``model`` argument corresponds to an existing local path, vLLM will load the config file directly from this path.
- If the ``model`` argument is a HuggingFace model ID consisting of a username and model name, vLLM will first try to use the config file from the HuggingFace local cache, using the ``model`` argument as the model name and the ``--revision`` argument as the revision. See `their website <https://huggingface.co/docs/huggingface_hub/en/package_reference/environment_variables#hfhome>`__ for more information on how the HuggingFace cache works.
- If the ``model`` argument is a HuggingFace model ID but it is not found in the cache, vLLM will download the config file from the HuggingFace model hub. Refer to `this function <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L91>`__ for the implementation. The input arguments include the ``model`` argument as the model name, the ``--revision`` argument as the revision, and the environment variable ``HF_TOKEN`` as the token to access the model hub. In our case, vLLM will download the `config.json <https://huggingface.co/Qwen/Qwen2-7B/blob/main/config.json>`__ file.
2. After confirming the existence of the model, vLLM loads its config file and converts it into a dictionary. See this `code snippet <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L185-L186>`__ for the implementation.
3. Next, vLLM `inspects <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L189>`__ the ``model_type`` field in the config dictionary to `generate <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#190-L216>`__ the config object to use. There are some ``model_type`` values that vLLM directly supports; see `here <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/transformers_utils/config.py#L48>`__ for the list. If the ``model_type`` is not in the list, vLLM will use `AutoConfig.from_pretrained <https://huggingface.co/docs/transformers/en/model_doc/auto#transformers.AutoConfig.from_pretrained>`__ to load the config class, with ``model``, ``--revision``, and ``--trust_remote_code`` as the arguments. Please note that:
- HuggingFace also has its own logic to determine the config class to use. It will again use the ``model_type`` field to search for the class name in the transformers library; see `here <https://github.com/huggingface/transformers/tree/main/src/transformers/models>`__ for the list of supported models. If the ``model_type`` is not found, HuggingFace will use the ``auto_map`` field from the config JSON file to determine the class name. Specifically, it is the ``AutoConfig`` field under ``auto_map``. See `DeepSeek <https://huggingface.co/deepseek-ai/DeepSeek-V2.5/blob/main/config.json>`__ for an example.
- The ``AutoConfig`` field under ``auto_map`` points to a module path in the model's repository. To create the config class, HuggingFace will import the module and use the ``from_pretrained`` method to load the config class. This can generally cause arbitrary code execution, so it is only executed when ``--trust_remote_code`` is enabled.
4. Subsequently, vLLM applies some historical patches to the config object. These are mostly related to RoPE configuration; see `here <https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/transformers_utils/config.py#L244>`__ for the implementation.
5. Finally, vLLM can reach the model class we want to initialize. vLLM uses the ``architectures`` field in the config object to determine the model class to initialize, as it maintains the mapping from architecture name to model class in `its registry <https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/model_executor/models/registry.py#L80>`__. If the architecture name is not found in the registry, it means this model architecture is not supported by vLLM. For ``Qwen/Qwen2-7B``, the ``architectures`` field is ``["Qwen2ForCausalLM"]``, which corresponds to the ``Qwen2ForCausalLM`` class in `vLLM's code <https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/model_executor/models/qwen2.py#L364>`__. This class will initialize itself depending on various configs.
Beyond that, there are two more things vLLM depends on HuggingFace for.
1. **Tokenizer**: vLLM uses the tokenizer from HuggingFace to tokenize the input text. The tokenizer is loaded using `AutoTokenizer.from_pretrained <https://huggingface.co/docs/transformers/en/model_doc/auto#transformers.AutoTokenizer.from_pretrained>`__ with the ``model`` argument as the model name and the ``--revision`` argument as the revision. It is also possible to use a tokenizer from another model by specifying the ``--tokenizer`` argument in the ``vllm serve`` command. Other relevant arguments are ``--tokenizer-revision`` and ``--tokenizer-mode``. Please check HuggingFace's documentation for the meaning of these arguments. This part of the logic can be found in the `get_tokenizer <https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/transformers_utils/tokenizer.py#L87>`__ function. After obtaining the tokenizer, notably, vLLM will cache some expensive attributes of the tokenizer in `get_cached_tokenizer <https://github.com/vllm-project/vllm/blob/127c07480ecea15e4c2990820c457807ff78a057/vllm/transformers_utils/tokenizer.py#L24>`__.
2. **Model weight**: vLLM downloads the model weight from the HuggingFace model hub using the ``model`` argument as the model name and the ``--revision`` argument as the revision. vLLM provides the argument ``--load-format`` to control what files to download from the model hub. By default, it will try to load the weights in the safetensors format and fall back to the PyTorch bin format if the safetensors format is not available. We can also pass ``--load-format dummy`` to skip downloading the weights.
- It is recommended to use the safetensors format, as it is efficient for loading in distributed inference and also safe from arbitrary code execution. See the `documentation <https://huggingface.co/docs/safetensors/en/index>`__ for more information on the safetensors format. This part of the logic can be found `here <https://github.com/vllm-project/vllm/blob/10b67d865d92e376956345becafc249d4c3c0ab7/vllm/model_executor/model_loader/loader.py#L385>`__. Please note that:
This completes the integration between vLLM and HuggingFace.
In summary, vLLM reads the config file ``config.json``, tokenizer, and model weight from the HuggingFace model hub or a local directory. It uses the config class from either vLLM, HuggingFace transformers, or loads the config class from the model's repository.

19
docs/source/design/input_processing/input_processing_pipeline.md Normal file
View File

@ -0,0 +1,19 @@
(input-processing-pipeline)=
# Input Processing Pipeline
1. Input data is passed to {class}`~vllm.LLMEngine` (or {class}`~vllm.AsyncLLMEngine`).
2. Tokenize the data if necessary.
3. Process the inputs using {meth}`INPUT_REGISTRY.process_input <vllm.inputs.registry.InputRegistry.process_input>`.
- For example, add placeholder tokens to reserve KV cache for multi-modal embeddings.
4. Send the processed inputs to {class}`~vllm.executor.executor_base.ExecutorBase`.
5. Distribute the inputs via {class}`~vllm.worker.worker_base.WorkerBase` to {class}`~vllm.worker.model_runner_base.ModelRunnerBase`.
6. If the data contains multi-modal data, convert it into keyword arguments using {meth}`MULTIMODAL_REGISTRY.map_input <vllm.multimodal.MultiModalRegistry.map_input>`.
- For example, convert a {class}`PIL.Image.Image` input to its pixel values for a vision model.

20
docs/source/design/input_processing/input_processing_pipeline.rst
View File

@ -1,20 +0,0 @@
.. _input_processing_pipeline:
Input Processing Pipeline
=========================
1. Input data is passed to :class:`~vllm.LLMEngine` (or :class:`~vllm.AsyncLLMEngine`).
2. Tokenize the data if necessary.
3. Process the inputs using :meth:`INPUT_REGISTRY.process_input <vllm.inputs.registry.InputRegistry.process_input>`.
- For example, add placeholder tokens to reserve KV cache for multi-modal embeddings.
4. Send the processed inputs to :class:`~vllm.executor.executor_base.ExecutorBase`.
5. Distribute the inputs via :class:`~vllm.worker.worker_base.WorkerBase` to :class:`~vllm.worker.model_runner_base.ModelRunnerBase`.
6. If the data contains multi-modal data, convert it into keyword arguments using :meth:`MULTIMODAL_REGISTRY.map_input <vllm.multimodal.MultiModalRegistry.map_input>`.
- For example, convert a :class:`PIL.Image.Image` input to its pixel values for a vision model.

43
docs/source/design/input_processing/model_inputs_index.md Normal file
View File

@ -0,0 +1,43 @@
(input-processing)=
# Input Processing
```{eval-rst}
.. currentmodule:: vllm.inputs
```
Each model can override parts of vLLM's [input processing pipeline](#input-processing-pipeline) via
{data}`~vllm.inputs.INPUT_REGISTRY` and {data}`~vllm.multimodal.MULTIMODAL_REGISTRY`.
Currently, this mechanism is only utilized in [multi-modal](#multi-modality) models for preprocessing multi-modal input
data in addition to input prompt, but it can be extended to text-only language models when needed.
## Guides
```{toctree}
:maxdepth: 1
input_processing_pipeline
```
## Module Contents
### LLM Engine Inputs
```{eval-rst}
.. autoclass:: vllm.inputs.DecoderOnlyInputs
:members:
:show-inheritance:
```
### Registry
```{eval-rst}
.. autodata:: vllm.inputs.INPUT_REGISTRY
```
```{eval-rst}
.. automodule:: vllm.inputs.registry
:members:
:show-inheritance:
```

39
docs/source/design/input_processing/model_inputs_index.rst
View File

@ -1,39 +0,0 @@
.. _input_processing:
Input Processing
================
.. currentmodule:: vllm.inputs
Each model can override parts of vLLM's :ref:`input processing pipeline <input_processing_pipeline>` via
:data:`~vllm.inputs.INPUT_REGISTRY` and :data:`~vllm.multimodal.MULTIMODAL_REGISTRY`.
Currently, this mechanism is only utilized in :ref:`multi-modal <multi_modality>` models for preprocessing multi-modal input
data in addition to input prompt, but it can be extended to text-only language models when needed.
Guides
++++++
.. toctree::
:maxdepth: 1
input_processing_pipeline
Module Contents
+++++++++++++++
LLM Engine Inputs
-----------------
.. autoclass:: vllm.inputs.DecoderOnlyInputs
:members:
:show-inheritance:
Registry
--------
.. autodata:: vllm.inputs.INPUT_REGISTRY
.. automodule:: vllm.inputs.registry
:members:
:show-inheritance:

527
docs/source/design/kernel/paged_attention.md Normal file
View File

@ -0,0 +1,527 @@
# vLLM Paged Attention
- Currently, vLLM utilizes its own implementation of a multi-head query
attention kernel (`csrc/attention/attention_kernels.cu`).
This kernel is designed to be compatible with
vLLM's paged KV caches, where the key and value cache are stored in
separate blocks (note that this block concept differs from the GPU
thread block. So in a later document, I will refer to vLLM paged
attention block as "block", while refer to GPU thread block as
"thread block").
- To achieve high performance, this kernel relies on a specially
designed memory layout and access method, specifically when threads
read data from global memory to shared memory. The purpose of this
document is to provide a high-level explanation of the kernel
implementation step by step, aiding those who wish to learn about the
vLLM multi-head query attention kernel. After going through this
document, users will likely have a better understanding and feel easier
to follow the actual implementation.
- Please note that this document may not cover all details, such as how
to calculate the correct index for the corresponding data or the dot
multiplication implementation. However, after reading this document
and becoming familiar with the high-level logic flow, it should be
easier for you to read the actual code and understand the details.
## Inputs
- The kernel function takes a list of arguments for the current thread
to perform its assigned work. The three most important arguments are
the input pointers `q`, `k_cache`, and `v_cache`, which point
to query, key, and value data on global memory that need to be read
and processed. The output pointer `out` points to global memory
where the result should be written. These four pointers actually
refer to multi-dimensional arrays, but each thread only accesses the
portion of data assigned to it. I have omitted all other runtime
parameters here for simplicity.
```cpp
template<
typename scalar_t,
int HEAD_SIZE,
int BLOCK_SIZE,
int NUM_THREADS,
int PARTITION_SIZE = 0>
__device__ void paged_attention_kernel(
... // Other side args.
const scalar_t* __restrict__ out, // [num_seqs, num_heads, max_num_partitions, head_size]
const scalar_t* __restrict__ q, // [num_seqs, num_heads, head_size]
const scalar_t* __restrict__ k_cache, // [num_blocks, num_kv_heads, head_size/x, block_size, x]
const scalar_t* __restrict__ v_cache, // [num_blocks, num_kv_heads, head_size, block_size]
... // Other side args.
)
```
- There are also a list of template arguments above the function
signature that are determined during compilation time. `scalar_t`
represents the data type of the query, key, and value data elements,
such as FP16. `HEAD_SIZE` indicates the number of elements in each
head. `BLOCK_SIZE` refers to the number of tokens in each block.
`NUM_THREADS` denotes the number of threads in each thread block.
`PARTITION_SIZE` represents the number of tensor parallel GPUs (For
simplicity, we assume this is 0 and tensor parallel is disabled).
- With these arguments, we need to perform a sequence of preparations.
This includes calculating the current head index, block index, and
other necessary variables. However, for now, we can ignore these
preparations and proceed directly to the actual calculations. It will
be easier to understand them once we grasp the entire flow.
## Concepts
- Just before we dive into the calculation flow, I want to describe a
few concepts that are needed for later sections. However, you may
skip this section and return later if you encounter any confusing
terminologies.
- **Sequence**: A sequence represents a client request. For example,
the data pointed to by `q` has a shape of
`[num_seqs, num_heads, head_size]`. That represents there are total
`num_seqs` of query sequence data are pointed by `q`. Since this
kernel is a single query attention kernel, each sequence only has one
query token. Hence, the `num_seqs` equals the total number of tokens
that are processed in the batch.
- **Context**: The context consists of the generated tokens from the
sequence. For instance, `["What", "is", "your"]` are the context
tokens, and the input query token is `"name"`. The model might
generate the token `"?"`.
- **Vec**: The vec is a list of elements that are fetched and
calculated together. For query and key data, the vec size
(`VEC_SIZE`) is determined so that each thread group can fetch and
calculate 16 bytes of data at a time. For value data, the vec size
(`V_VEC_SIZE`) is determined so that each thread can fetch and
calculate 16 bytes of data at a time. For example, if the
`scalar_t` is FP16 (2 bytes) and `THREAD_GROUP_SIZE` is 2, the
`VEC_SIZE` will be 4, while the `V_VEC_SIZE` will be 8.
- **Thread group**: The thread group is a small group of
threads(`THREAD_GROUP_SIZE`) that fetches and calculates one
query token and one key token at a time. Each thread handles only a
portion of the token data. The total number of elements processed by
one thread group is referred as `x`. For example, if the thread
group contains 2 threads and the head size is 8, then thread 0
handles the query and key elements at index 0, 2, 4, 6, while thread
1 handles the elements at index 1, 3, 5, 7.
- **Block**: The key and value cache data in vLLM are split into
blocks. Each block stores data for a fixed number(`BLOCK_SIZE`)
of tokens at one head. Each block may contain only a portion of the
whole context tokens. For example, if the block size is 16 and the
head size is 128, then for one head, one block can store 16 * 128 =
2048 elements.
- **Warp**: A warp is a group of 32 threads(`WARP_SIZE`) that
execute simultaneously on a stream multiprocessor (SM). In this
kernel, each warp processes the calculation between one query token
and key tokens of one entire block at a time (it may process multiple
blocks in multiple iterations). For example, if there are 4 warps and
6 blocks for one context, the assignment would be like warp 0 handles
the 0th, 4th blocks, warp 1 handles the 1st, 5th blocks, warp 2
handles the 2nd block and warp 3 handles the 3rd block.
- **Thread block**: A thread block is a group of
threads(`NUM_THREADS`) that can access the same shared memory.
Each thread block contains multiple warps(`NUM_WARPS`), and in
this kernel, each thread block processes the calculation between one
query token and key tokens of a whole context.
- **Grid**: A grid is a collection of thread blocks and defines the
shape of the collection. In this kernel, the shape is
`(num_heads, num_seqs, max_num_partitions)`. Therefore, each thread
block only handles the calculation for one head, one sequence, and
one partition.
## Query
- This section will introduce how query data is stored in memory and
fetched by each thread. As mentioned above, each thread group fetches
one query token data, while each thread itself only handles a part of
one query token data. Within each warp, every thread group will fetch
the same query token data, but will multiply it with different key
token data.
```cpp
const scalar_t* q_ptr = q + seq_idx * q_stride + head_idx * HEAD_SIZE;
```
```{figure} ../../assets/kernel/query.png
:align: center
:alt: query
:width: 70%
Query data of one token at one head
```
- Each thread defines its own `q_ptr` which points to the assigned
query token data on global memory. For example, if `VEC_SIZE` is 4
and `HEAD_SIZE` is 128, the `q_ptr` points to data that contains
total of 128 elements divided into 128 / 4 = 32 vecs.
```{figure} ../../assets/kernel/q_vecs.png
:align: center
:alt: q_vecs
:width: 70%
`q_vecs` for one thread group
```
```cpp
__shared__ Q_vec q_vecs[THREAD_GROUP_SIZE][NUM_VECS_PER_THREAD];
```
- Next, we need to read the global memory data pointed to by `q_ptr`
into shared memory as `q_vecs`. It is important to note that each
vecs is assigned to a different row. For example, if the
`THREAD_GROUP_SIZE` is 2, thread 0 will handle the 0th row vecs,
while thread 1 handles the 1st row vecs. By reading the query data in
this way, neighboring threads like thread 0 and thread 1 can read
neighbor memory, achieving the memory coalescing to improve
performance.
## Key
- Similar to the "Query" section, this section introduces memory layout
and assignment for keys. While each thread group only handle one
query token one kernel run, it may handle multiple key tokens across
multiple iterations. Meanwhile, each warp will process multiple blocks
of key tokens in multiple iterations, ensuring that all context
tokens are processed by the entire thread group after the kernel run.
In this context, "handle" refers to performing the dot multiplication
between query data and key data.
```cpp
const scalar_t* k_ptr = k_cache + physical_block_number * kv_block_stride
+ kv_head_idx * kv_head_stride
+ physical_block_offset * x;
```
- Unlike to `q_ptr`, `k_ptr` in each thread will point to different
key token at different iterations. As shown above, that `k_ptr`
points to key token data based on `k_cache` at assigned block,
assigned head and assigned token.
```{figure} ../../assets/kernel/key.png
:align: center
:alt: key
:width: 70%
Key data of all context tokens at one head
```
- The diagram above illustrates the memory layout for key data. It
assumes that the `BLOCK_SIZE` is 16, `HEAD_SIZE` is 128, `x` is
8, `THREAD_GROUP_SIZE` is 2, and there are a total of 4 warps. Each
rectangle represents all the elements for one key token at one head,
which will be processed by one thread group. The left half shows the
total 16 blocks of key token data for warp 0, while the right half
represents the remaining key token data for other warps or
iterations. Inside each rectangle, there are a total 32 vecs (128
elements for one token) that will be processed by 2 threads (one
thread group) separately.
```{figure} ../../assets/kernel/k_vecs.png
:align: center
:alt: k_vecs
:width: 70%
`k_vecs` for one thread
```
```cpp
K_vec k_vecs[NUM_VECS_PER_THREAD]
```
- Next, we need to read the key token data from `k_ptr` and store
them on register memory as `k_vecs`. We use register memory for
`k_vecs` because it will only be accessed by one thread once,
whereas `q_vecs` will be accessed by multiple threads multiple
times. Each `k_vecs` will contain multiple vectors for later
calculation. Each vec will be set at each inner iteration. The
assignment of vecs allows neighboring threads in a warp to read
neighboring memory together, which again promotes the memory
coalescing. For instance, thread 0 will read vec 0, while thread 1
will read vec 1. In the next inner loop, thread 0 will read vec 2,
while thread 1 will read vec 3, and so on.
- You may still be a little confused about the overall flow. Don't
worry, please keep reading the next "QK" section. It will illustrate
the query and key calculation flow in a clearer and higher-level
manner.
## QK
- As shown the pseudo code below, before the entire for loop block, we
fetch the query data for one token and store it in `q_vecs`. Then,
in the outer for loop, we iterate through different `k_ptrs` that
point to different tokens and prepare the `k_vecs` in the inner for
loop. Finally, we perform the dot multiplication between the
`q_vecs` and each `k_vecs`.
```cpp
q_vecs = ...
for ... {
k_ptr = ...
for ... {
k_vecs[i] = ...
}
...
float qk = scale * Qk_dot<scalar_t, THREAD_GROUP_SIZE>::dot(q_vecs[thread_group_offset], k_vecs);
}
```
- As mentioned before, for each thread, it only fetches part of the
query and key token data at a time. However, there will be a cross
thread group reduction happen in the `Qk_dot<>::dot` . So `qk`
returned here is not just between part of the query and key token dot
multiplication, but actually a full result between entire query and
key token data.
- For example, if the value of `HEAD_SIZE` is 128 and
`THREAD_GROUP_SIZE` is 2, each thread's `k_vecs` will contain
total 64 elements. However, the returned `qk` is actually the
result of dot multiplication between 128 query elements and 128 key
elements. If you want to learn more about the details of the dot
multiplication and reduction, you may refer to the implementation of
`Qk_dot<>::dot`. However, for the sake of simplicity, I will not
cover it in this document.
## Softmax
- Next, we need to calculate the normalized softmax for all `qk`s,
as shown above, where each $x$ represents a `qk`. To do this,
we must obtain the reduced value of `qk_max`($m(x)$) and
the `exp_sum`($\ell(x)$) of all `qk`s. The reduction
should be performed across the entire thread block, encompassing
results between the query token and all context key tokens.
```{math}
:nowrap: true
\begin{gather*}
m(x):=\max _i \quad x_i \\ \quad f(x):=\left[\begin{array}{lll}e^{x_1-m(x)} & \ldots & e^{x_B-m(x)}\end{array}\right]\\ \quad \ell(x):=\sum_i f(x)_i \\
\quad \operatorname{softmax}(x):=\frac{f(x)}{\ell(x)}
\end{gather*}
```
### `qk_max` and `logits`
- Just right after we get the `qk` result, we can set the temporary
`logits` result with `qk` (In the end, the `logits` should
store the normalized softmax result). Also we can compare and collect
the `qk_max` for all `qk`s that are calculated by current
thread group.
```cpp
if (thread_group_offset == 0) {
const bool mask = token_idx >= context_len;
logits[token_idx - start_token_idx] = mask ? 0.f : qk;
qk_max = mask ? qk_max : fmaxf(qk_max, qk);
}
```
- Please note that the `logits` here is on shared memory, so each
thread group will set the fields for its own assigned context tokens.
Overall, the size of logits should be number of context tokens.
```cpp
for (int mask = WARP_SIZE / 2; mask >= THREAD_GROUP_SIZE; mask /= 2) {
qk_max = fmaxf(qk_max, VLLM_SHFL_XOR_SYNC(qk_max, mask));
}
if (lane == 0) {
red_smem[warp_idx] = qk_max;
}
```
- Then we need to get the reduced `qk_max` across each warp. The main
idea is to make threads in warp to communicate with each other and
get the final max `qk` .
```cpp
for (int mask = NUM_WARPS / 2; mask >= 1; mask /= 2) {
qk_max = fmaxf(qk_max, VLLM_SHFL_XOR_SYNC(qk_max, mask));
}
qk_max = VLLM_SHFL_SYNC(qk_max, 0);
```
- Finally, we can get the reduced `qk_max` from whole thread block by
compare the `qk_max` from all warps in this thread block. Then we
need to broadcast the final result to each thread.
### `exp_sum`
- Similar to `qk_max`, we need to get the reduced sum value from the
entire thread block too.
```cpp
for (int i = thread_idx; i < num_tokens; i += NUM_THREADS) {
float val = __expf(logits[i] - qk_max);
logits[i] = val;
exp_sum += val;
}
...
exp_sum = block_sum<NUM_WARPS>(&red_smem[NUM_WARPS], exp_sum);
```
- Firstly, sum all exp values from each thread group, and meanwhile,
convert each entry of `logits` from `qk` to `exp(qk - qk_max)`.
Please note, the `qk_max` here is already the max `qk` across the
whole thread block. And then we can do reduction for `exp_sum`
across whole thread block just like the `qk_max`.
```cpp
const float inv_sum = __fdividef(1.f, exp_sum + 1e-6f);
for (int i = thread_idx; i < num_tokens; i += NUM_THREADS) {
logits[i] *= inv_sum;
}
```
- Finally, with the reduced `qk_max` and `exp_sum`, we can obtain
the final normalized softmax result as `logits`. This `logits`
variable will be used for dot multiplication with the value data in
later steps. Now, it should store the normalized softmax result of
`qk` for all assigned context tokens.
## Value
```{figure} ../../assets/kernel/value.png
:align: center
:alt: value
:width: 70%
Value data of all context tokens at one head
```
```{figure} ../../assets/kernel/logits_vec.png
:align: center
:alt: logits_vec
:width: 50%
`logits_vec` for one thread
```
```{figure} ../../assets/kernel/v_vec.png
:align: center
:alt: v_vec
:width: 70%
List of `v_vec` for one thread
```
- Now we need to retrieve the value data and perform dot multiplication
with `logits`. Unlike query and key, there is no thread group
concept for value data. As shown in diagram, different from key token
memory layout, elements from the same column correspond to the same
value token. For one block of value data, there are `HEAD_SIZE` of
rows and `BLOCK_SIZE` of columns that are split into multiple
`v_vecs`.
- Each thread always fetches `V_VEC_SIZE` elements from the same
`V_VEC_SIZE` of tokens at a time. As a result, a single thread
retrieves multiple `v_vec`s from different rows and the same
columns through multiple inner iterations. For each `v_vec`, it
needs to be dot multiplied with the corresponding `logits_vec`,
which is also `V_VEC_SIZE` elements from `logits`. Overall, with
multiple inner iterations, each warp will process one block of value
tokens. And with multiple outer iterations, the whole context value
tokens are processd
```cpp
float accs[NUM_ROWS_PER_THREAD];
for ... { // Iteration over different blocks.
logits_vec = ...
for ... { // Iteration over different rows.
v_vec = ...
...
accs[i] += dot(logits_vec, v_vec);
}
}
```
- As shown in the above pseudo code, in the outer loop, similar to
`k_ptr`, `logits_vec` iterates over different blocks and reads
`V_VEC_SIZE` elements from `logits`. In the inner loop, each
thread reads `V_VEC_SIZE` elements from the same tokens as a
`v_vec` and performs dot multiplication. It is important to note
that in each inner iteration, the thread fetches different head
position elements for the same tokens. The dot result is then
accumulated in `accs`. Therefore, each entry of `accs` is mapped
to a head position assigned to the current thread.
- For example, if `BLOCK_SIZE` is 16 and `V_VEC_SIZE` is 8, each
thread fetches 8 value elements for 8 tokens at a time. Each element
is from different tokens at the same head position. If `HEAD_SIZE`
is 128 and `WARP_SIZE` is 32, for each inner loop, a warp needs to
fetch `WARP_SIZE * V_VEC_SIZE = 256` elements. This means there are
a total of 128 * 16 / 256 = 8 inner iterations for a warp to handle
a whole block of value tokens. And each `accs` in each thread
contains 8 elements that accumulated at 8 different head positions.
For the thread 0, the `accs` variable will have 8 elements, which
are 0th, 32th … 224th elements of a value head that are accumulated
from all assigned 8 tokens.
## LV
- Now, we need to perform reduction for `accs` within each warp. This
process allows each thread to accumulate the `accs` for the
assigned head positions of all tokens in one block.
```cpp
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
float acc = accs[i];
for (int mask = NUM_V_VECS_PER_ROW / 2; mask >= 1; mask /= 2) {
acc += VLLM_SHFL_XOR_SYNC(acc, mask);
}
accs[i] = acc;
}
```
- Next, we perform reduction for `accs` across all warps, allowing
each thread to have the accumulation of `accs` for the assigned
head positions of all context tokens. Please note that each `accs`
in every thread only stores the accumulation for a portion of
elements of the entire head for all context tokens. However, overall,
all results for output have been calculated but are just stored in
different thread register memory.
```cpp
float* out_smem = reinterpret_cast<float*>(shared_mem);
for (int i = NUM_WARPS; i > 1; i /= 2) {
// Upper warps write to shared memory.
...
float* dst = &out_smem[(warp_idx - mid) * HEAD_SIZE];
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
...
dst[row_idx] = accs[i];
}
// Lower warps update the output.
const float* src = &out_smem[warp_idx * HEAD_SIZE];
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
...
accs[i] += src[row_idx];
}
// Write out the accs.
}
```
## Output
- Now we can write all of calculated result from local register memory
to final output global memory.
```cpp
scalar_t* out_ptr = out + seq_idx * num_heads * max_num_partitions * HEAD_SIZE
+ head_idx * max_num_partitions * HEAD_SIZE
+ partition_idx * HEAD_SIZE;
```
- First, we need to define the `out_ptr` variable, which points to
the start address of the assigned sequence and assigned head.
```cpp
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
const int row_idx = lane / NUM_V_VECS_PER_ROW + i * NUM_ROWS_PER_ITER;
if (row_idx < HEAD_SIZE && lane % NUM_V_VECS_PER_ROW == 0) {
from_float(*(out_ptr + row_idx), accs[i]);
}
}
```
- Finally, we need to iterate over different assigned head positions
and write out the corresponding accumulated result based on the
`out_ptr`.

525
docs/source/design/kernel/paged_attention.rst
View File

@ -1,525 +0,0 @@
vLLM Paged Attention
====================
- Currently, vLLM utilizes its own implementation of a multi-head query
attention kernel (``csrc/attention/attention_kernels.cu``).
This kernel is designed to be compatible with
vLLM's paged KV caches, where the key and value cache are stored in
separate blocks (note that this block concept differs from the GPU
thread block. So in a later document, I will refer to vLLM paged
attention block as "block", while refer to GPU thread block as
"thread block").
- To achieve high performance, this kernel relies on a specially
designed memory layout and access method, specifically when threads
read data from global memory to shared memory. The purpose of this
document is to provide a high-level explanation of the kernel
implementation step by step, aiding those who wish to learn about the
vLLM multi-head query attention kernel. After going through this
document, users will likely have a better understanding and feel easier
to follow the actual implementation.
- Please note that this document may not cover all details, such as how
to calculate the correct index for the corresponding data or the dot
multiplication implementation. However, after reading this document
and becoming familiar with the high-level logic flow, it should be
easier for you to read the actual code and understand the details.
Inputs
------
- The kernel function takes a list of arguments for the current thread
to perform its assigned work. The three most important arguments are
the input pointers ``q``, ``k_cache``, and ``v_cache``, which point
to query, key, and value data on global memory that need to be read
and processed. The output pointer ``out`` points to global memory
where the result should be written. These four pointers actually
refer to multi-dimensional arrays, but each thread only accesses the
portion of data assigned to it. I have omitted all other runtime
parameters here for simplicity.
.. code:: cpp
template<
typename scalar_t,
int HEAD_SIZE,
int BLOCK_SIZE,
int NUM_THREADS,
int PARTITION_SIZE = 0>
__device__ void paged_attention_kernel(
... // Other side args.
const scalar_t* __restrict__ out, // [num_seqs, num_heads, max_num_partitions, head_size]
const scalar_t* __restrict__ q, // [num_seqs, num_heads, head_size]
const scalar_t* __restrict__ k_cache, // [num_blocks, num_kv_heads, head_size/x, block_size, x]
const scalar_t* __restrict__ v_cache, // [num_blocks, num_kv_heads, head_size, block_size]
... // Other side args.
)
- There are also a list of template arguments above the function
signature that are determined during compilation time. ``scalar_t``
represents the data type of the query, key, and value data elements,
such as FP16. ``HEAD_SIZE`` indicates the number of elements in each
head. ``BLOCK_SIZE`` refers to the number of tokens in each block.
``NUM_THREADS`` denotes the number of threads in each thread block.
``PARTITION_SIZE`` represents the number of tensor parallel GPUs (For
simplicity, we assume this is 0 and tensor parallel is disabled).
- With these arguments, we need to perform a sequence of preparations.
This includes calculating the current head index, block index, and
other necessary variables. However, for now, we can ignore these
preparations and proceed directly to the actual calculations. It will
be easier to understand them once we grasp the entire flow.
Concepts
--------
- Just before we dive into the calculation flow, I want to describe a
few concepts that are needed for later sections. However, you may
skip this section and return later if you encounter any confusing
terminologies.
- **Sequence**: A sequence represents a client request. For example,
the data pointed to by ``q`` has a shape of
``[num_seqs, num_heads, head_size]``. That represents there are total
``num_seqs`` of query sequence data are pointed by ``q``. Since this
kernel is a single query attention kernel, each sequence only has one
query token. Hence, the ``num_seqs`` equals the total number of tokens
that are processed in the batch.
- **Context**: The context consists of the generated tokens from the
sequence. For instance, ``["What", "is", "your"]`` are the context
tokens, and the input query token is ``"name"``. The model might
generate the token ``"?"``.
- **Vec**: The vec is a list of elements that are fetched and
calculated together. For query and key data, the vec size
(``VEC_SIZE``) is determined so that each thread group can fetch and
calculate 16 bytes of data at a time. For value data, the vec size
(``V_VEC_SIZE``) is determined so that each thread can fetch and
calculate 16 bytes of data at a time. For example, if the
``scalar_t`` is FP16 (2 bytes) and ``THREAD_GROUP_SIZE`` is 2, the
``VEC_SIZE`` will be 4, while the ``V_VEC_SIZE`` will be 8.
- **Thread group**: The thread group is a small group of
threads(\ ``THREAD_GROUP_SIZE``) that fetches and calculates one
query token and one key token at a time. Each thread handles only a
portion of the token data. The total number of elements processed by
one thread group is referred as ``x``. For example, if the thread
group contains 2 threads and the head size is 8, then thread 0
handles the query and key elements at index 0, 2, 4, 6, while thread
1 handles the elements at index 1, 3, 5, 7.
- **Block**: The key and value cache data in vLLM are split into
blocks. Each block stores data for a fixed number(\ ``BLOCK_SIZE``)
of tokens at one head. Each block may contain only a portion of the
whole context tokens. For example, if the block size is 16 and the
head size is 128, then for one head, one block can store 16 \* 128 =
2048 elements.
- **Warp**: A warp is a group of 32 threads(\ ``WARP_SIZE``) that
execute simultaneously on a stream multiprocessor (SM). In this
kernel, each warp processes the calculation between one query token
and key tokens of one entire block at a time (it may process multiple
blocks in multiple iterations). For example, if there are 4 warps and
6 blocks for one context, the assignment would be like warp 0 handles
the 0th, 4th blocks, warp 1 handles the 1st, 5th blocks, warp 2
handles the 2nd block and warp 3 handles the 3rd block.
- **Thread block**: A thread block is a group of
threads(\ ``NUM_THREADS``) that can access the same shared memory.
Each thread block contains multiple warps(\ ``NUM_WARPS``), and in
this kernel, each thread block processes the calculation between one
query token and key tokens of a whole context.
- **Grid**: A grid is a collection of thread blocks and defines the
shape of the collection. In this kernel, the shape is
``(num_heads, num_seqs, max_num_partitions)``. Therefore, each thread
block only handles the calculation for one head, one sequence, and
one partition.
Query
-----
- This section will introduce how query data is stored in memory and
fetched by each thread. As mentioned above, each thread group fetches
one query token data, while each thread itself only handles a part of
one query token data. Within each warp, every thread group will fetch
the same query token data, but will multiply it with different key
token data.
.. code:: cpp
const scalar_t* q_ptr = q + seq_idx * q_stride + head_idx * HEAD_SIZE;
.. figure:: ../../assets/kernel/query.png
:alt: query
:width: 70%
:align: center
Query data of one token at one head
- Each thread defines its own ``q_ptr`` which points to the assigned
query token data on global memory. For example, if ``VEC_SIZE`` is 4
and ``HEAD_SIZE`` is 128, the ``q_ptr`` points to data that contains
total of 128 elements divided into 128 / 4 = 32 vecs.
.. figure:: ../../assets/kernel/q_vecs.png
:alt: q_vecs
:width: 70%
:align: center
``q_vecs`` for one thread group
.. code:: cpp
__shared__ Q_vec q_vecs[THREAD_GROUP_SIZE][NUM_VECS_PER_THREAD];
- Next, we need to read the global memory data pointed to by ``q_ptr``
into shared memory as ``q_vecs``. It is important to note that each
vecs is assigned to a different row. For example, if the
``THREAD_GROUP_SIZE`` is 2, thread 0 will handle the 0th row vecs,
while thread 1 handles the 1st row vecs. By reading the query data in
this way, neighboring threads like thread 0 and thread 1 can read
neighbor memory, achieving the memory coalescing to improve
performance.
Key
---
- Similar to the "Query" section, this section introduces memory layout
and assignment for keys. While each thread group only handle one
query token one kernel run, it may handle multiple key tokens across
multiple iterations. Meanwhile, each warp will process multiple blocks
of key tokens in multiple iterations, ensuring that all context
tokens are processed by the entire thread group after the kernel run.
In this context, "handle" refers to performing the dot multiplication
between query data and key data.
.. code:: cpp
const scalar_t* k_ptr = k_cache + physical_block_number * kv_block_stride
+ kv_head_idx * kv_head_stride
+ physical_block_offset * x;
- Unlike to ``q_ptr``, ``k_ptr`` in each thread will point to different
key token at different iterations. As shown above, that ``k_ptr``
points to key token data based on ``k_cache`` at assigned block,
assigned head and assigned token.
.. figure:: ../../assets/kernel/key.png
:alt: key
:width: 70%
:align: center
Key data of all context tokens at one head
- The diagram above illustrates the memory layout for key data. It
assumes that the ``BLOCK_SIZE`` is 16, ``HEAD_SIZE`` is 128, ``x`` is
8, ``THREAD_GROUP_SIZE`` is 2, and there are a total of 4 warps. Each
rectangle represents all the elements for one key token at one head,
which will be processed by one thread group. The left half shows the
total 16 blocks of key token data for warp 0, while the right half
represents the remaining key token data for other warps or
iterations. Inside each rectangle, there are a total 32 vecs (128
elements for one token) that will be processed by 2 threads (one
thread group) separately.
.. figure:: ../../assets/kernel/k_vecs.png
:alt: k_vecs
:width: 70%
:align: center
``k_vecs`` for one thread
.. code:: cpp
K_vec k_vecs[NUM_VECS_PER_THREAD]
- Next, we need to read the key token data from ``k_ptr`` and store
them on register memory as ``k_vecs``. We use register memory for
``k_vecs`` because it will only be accessed by one thread once,
whereas ``q_vecs`` will be accessed by multiple threads multiple
times. Each ``k_vecs`` will contain multiple vectors for later
calculation. Each vec will be set at each inner iteration. The
assignment of vecs allows neighboring threads in a warp to read
neighboring memory together, which again promotes the memory
coalescing. For instance, thread 0 will read vec 0, while thread 1
will read vec 1. In the next inner loop, thread 0 will read vec 2,
while thread 1 will read vec 3, and so on.
- You may still be a little confused about the overall flow. Don't
worry, please keep reading the next "QK" section. It will illustrate
the query and key calculation flow in a clearer and higher-level
manner.
QK
---
- As shown the pseudo code below, before the entire for loop block, we
fetch the query data for one token and store it in ``q_vecs``. Then,
in the outer for loop, we iterate through different ``k_ptrs`` that
point to different tokens and prepare the ``k_vecs`` in the inner for
loop. Finally, we perform the dot multiplication between the
``q_vecs`` and each ``k_vecs``.
.. code:: cpp
q_vecs = ...
for ... {
k_ptr = ...
for ... {
k_vecs[i] = ...
}
...
float qk = scale * Qk_dot<scalar_t, THREAD_GROUP_SIZE>::dot(q_vecs[thread_group_offset], k_vecs);
}
- As mentioned before, for each thread, it only fetches part of the
query and key token data at a time. However, there will be a cross
thread group reduction happen in the ``Qk_dot<>::dot`` . So ``qk``
returned here is not just between part of the query and key token dot
multiplication, but actually a full result between entire query and
key token data.
- For example, if the value of ``HEAD_SIZE`` is 128 and
``THREAD_GROUP_SIZE`` is 2, each thread's ``k_vecs`` will contain
total 64 elements. However, the returned ``qk`` is actually the
result of dot multiplication between 128 query elements and 128 key
elements. If you want to learn more about the details of the dot
multiplication and reduction, you may refer to the implementation of
``Qk_dot<>::dot``. However, for the sake of simplicity, I will not
cover it in this document.
Softmax
-------
- Next, we need to calculate the normalized softmax for all ``qk``\ s,
as shown above, where each :math:`x` represents a ``qk``. To do this,
we must obtain the reduced value of ``qk_max``\ (:math:`m(x)`) and
the ``exp_sum``\ (:math:`\ell(x)`) of all ``qk``\ s. The reduction
should be performed across the entire thread block, encompassing
results between the query token and all context key tokens.
.. math::
:nowrap:
\begin{gather*}
m(x):=\max _i \quad x_i \\ \quad f(x):=\left[\begin{array}{lll}e^{x_1-m(x)} & \ldots & e^{x_B-m(x)}\end{array}\right]\\ \quad \ell(x):=\sum_i f(x)_i \\
\quad \operatorname{softmax}(x):=\frac{f(x)}{\ell(x)}
\end{gather*}
``qk_max`` and ``logits``
~~~~~~~~~~~~~~~~~~~~~~~~~
- Just right after we get the ``qk`` result, we can set the temporary
``logits`` result with ``qk`` (In the end, the ``logits`` should
store the normalized softmax result). Also we can compare and collect
the ``qk_max`` for all ``qk``\ s that are calculated by current
thread group.
.. code:: cpp
if (thread_group_offset == 0) {
const bool mask = token_idx >= context_len;
logits[token_idx - start_token_idx] = mask ? 0.f : qk;
qk_max = mask ? qk_max : fmaxf(qk_max, qk);
}
- Please note that the ``logits`` here is on shared memory, so each
thread group will set the fields for its own assigned context tokens.
Overall, the size of logits should be number of context tokens.
.. code:: cpp
for (int mask = WARP_SIZE / 2; mask >= THREAD_GROUP_SIZE; mask /= 2) {
qk_max = fmaxf(qk_max, VLLM_SHFL_XOR_SYNC(qk_max, mask));
}
if (lane == 0) {
red_smem[warp_idx] = qk_max;
}
- Then we need to get the reduced ``qk_max`` across each warp. The main
idea is to make threads in warp to communicate with each other and
get the final max ``qk`` .
.. code:: cpp
for (int mask = NUM_WARPS / 2; mask >= 1; mask /= 2) {
qk_max = fmaxf(qk_max, VLLM_SHFL_XOR_SYNC(qk_max, mask));
}
qk_max = VLLM_SHFL_SYNC(qk_max, 0);
- Finally, we can get the reduced ``qk_max`` from whole thread block by
compare the ``qk_max`` from all warps in this thread block. Then we
need to broadcast the final result to each thread.
``exp_sum``
~~~~~~~~~~~
- Similar to ``qk_max``, we need to get the reduced sum value from the
entire thread block too.
.. code:: cpp
for (int i = thread_idx; i < num_tokens; i += NUM_THREADS) {
float val = __expf(logits[i] - qk_max);
logits[i] = val;
exp_sum += val;
}
...
exp_sum = block_sum<NUM_WARPS>(&red_smem[NUM_WARPS], exp_sum);
- Firstly, sum all exp values from each thread group, and meanwhile,
convert each entry of ``logits`` from ``qk`` to ``exp(qk - qk_max)``.
Please note, the ``qk_max`` here is already the max ``qk`` across the
whole thread block. And then we can do reduction for ``exp_sum``
across whole thread block just like the ``qk_max``.
.. code:: cpp
const float inv_sum = __fdividef(1.f, exp_sum + 1e-6f);
for (int i = thread_idx; i < num_tokens; i += NUM_THREADS) {
logits[i] *= inv_sum;
}
- Finally, with the reduced ``qk_max`` and ``exp_sum``, we can obtain
the final normalized softmax result as ``logits``. This ``logits``
variable will be used for dot multiplication with the value data in
later steps. Now, it should store the normalized softmax result of
``qk`` for all assigned context tokens.
Value
-----
.. figure:: ../../assets/kernel/value.png
:alt: value
:width: 70%
:align: center
Value data of all context tokens at one head
.. figure:: ../../assets/kernel/logits_vec.png
:alt: logits_vec
:width: 50%
:align: center
``logits_vec`` for one thread
.. figure:: ../../assets/kernel/v_vec.png
:alt: v_vec
:width: 70%
:align: center
List of ``v_vec`` for one thread
- Now we need to retrieve the value data and perform dot multiplication
with ``logits``. Unlike query and key, there is no thread group
concept for value data. As shown in diagram, different from key token
memory layout, elements from the same column correspond to the same
value token. For one block of value data, there are ``HEAD_SIZE`` of
rows and ``BLOCK_SIZE`` of columns that are split into multiple
``v_vecs``.
- Each thread always fetches ``V_VEC_SIZE`` elements from the same
``V_VEC_SIZE`` of tokens at a time. As a result, a single thread
retrieves multiple ``v_vec``\ s from different rows and the same
columns through multiple inner iterations. For each ``v_vec``, it
needs to be dot multiplied with the corresponding ``logits_vec``,
which is also ``V_VEC_SIZE`` elements from ``logits``. Overall, with
multiple inner iterations, each warp will process one block of value
tokens. And with multiple outer iterations, the whole context value
tokens are processd
.. code:: cpp
float accs[NUM_ROWS_PER_THREAD];
for ... { // Iteration over different blocks.
logits_vec = ...
for ... { // Iteration over different rows.
v_vec = ...
...
accs[i] += dot(logits_vec, v_vec);
}
}
- As shown in the above pseudo code, in the outer loop, similar to
``k_ptr``, ``logits_vec`` iterates over different blocks and reads
``V_VEC_SIZE`` elements from ``logits``. In the inner loop, each
thread reads ``V_VEC_SIZE`` elements from the same tokens as a
``v_vec`` and performs dot multiplication. It is important to note
that in each inner iteration, the thread fetches different head
position elements for the same tokens. The dot result is then
accumulated in ``accs``. Therefore, each entry of ``accs`` is mapped
to a head position assigned to the current thread.
- For example, if ``BLOCK_SIZE`` is 16 and ``V_VEC_SIZE`` is 8, each
thread fetches 8 value elements for 8 tokens at a time. Each element
is from different tokens at the same head position. If ``HEAD_SIZE``
is 128 and ``WARP_SIZE`` is 32, for each inner loop, a warp needs to
fetch ``WARP_SIZE * V_VEC_SIZE = 256`` elements. This means there are
a total of 128 \* 16 / 256 = 8 inner iterations for a warp to handle
a whole block of value tokens. And each ``accs`` in each thread
contains 8 elements that accumulated at 8 different head positions.
For the thread 0, the ``accs`` variable will have 8 elements, which
are 0th, 32th … 224th elements of a value head that are accumulated
from all assigned 8 tokens.
LV
---
- Now, we need to perform reduction for ``accs`` within each warp. This
process allows each thread to accumulate the ``accs`` for the
assigned head positions of all tokens in one block.
.. code:: cpp
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
float acc = accs[i];
for (int mask = NUM_V_VECS_PER_ROW / 2; mask >= 1; mask /= 2) {
acc += VLLM_SHFL_XOR_SYNC(acc, mask);
}
accs[i] = acc;
}
- Next, we perform reduction for ``accs`` across all warps, allowing
each thread to have the accumulation of ``accs`` for the assigned
head positions of all context tokens. Please note that each ``accs``
in every thread only stores the accumulation for a portion of
elements of the entire head for all context tokens. However, overall,
all results for output have been calculated but are just stored in
different thread register memory.
.. code:: cpp
float* out_smem = reinterpret_cast<float*>(shared_mem);
for (int i = NUM_WARPS; i > 1; i /= 2) {
// Upper warps write to shared memory.
...
float* dst = &out_smem[(warp_idx - mid) * HEAD_SIZE];
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
...
dst[row_idx] = accs[i];
}
// Lower warps update the output.
const float* src = &out_smem[warp_idx * HEAD_SIZE];
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
...
accs[i] += src[row_idx];
}
// Write out the accs.
}
Output
------
- Now we can write all of calculated result from local register memory
to final output global memory.
.. code:: cpp
scalar_t* out_ptr = out + seq_idx * num_heads * max_num_partitions * HEAD_SIZE
+ head_idx * max_num_partitions * HEAD_SIZE
+ partition_idx * HEAD_SIZE;
- First, we need to define the ``out_ptr`` variable, which points to
the start address of the assigned sequence and assigned head.
.. code:: cpp
for (int i = 0; i < NUM_ROWS_PER_THREAD; i++) {
const int row_idx = lane / NUM_V_VECS_PER_ROW + i * NUM_ROWS_PER_ITER;
if (row_idx < HEAD_SIZE && lane % NUM_V_VECS_PER_ROW == 0) {
from_float(*(out_ptr + row_idx), accs[i]);
}
}
- Finally, we need to iterate over different assigned head positions
and write out the corresponding accumulated result based on the
``out_ptr``.

16
docs/source/design/multimodal/adding_multimodal_plugin.md Normal file
View File

@ -0,0 +1,16 @@
(adding-multimodal-plugin)=
# Adding a Multimodal Plugin
This document teaches you how to add a new modality to vLLM.
Each modality in vLLM is represented by a {class}`~vllm.multimodal.MultiModalPlugin` and registered to {data}`~vllm.multimodal.MULTIMODAL_REGISTRY`.
For vLLM to recognize a new modality type, you have to create a new plugin and then pass it to {meth}`~vllm.multimodal.MultiModalRegistry.register_plugin`.
The remainder of this document details how to define custom {class}`~vllm.multimodal.MultiModalPlugin` s.
```{note}
This article is a work in progress.
```
% TODO: Add more instructions on how to add new plugins once embeddings is in.

17
docs/source/design/multimodal/adding_multimodal_plugin.rst
View File

@ -1,17 +0,0 @@
.. _adding_multimodal_plugin:
Adding a Multimodal Plugin
==========================
This document teaches you how to add a new modality to vLLM.
Each modality in vLLM is represented by a :class:`~vllm.multimodal.MultiModalPlugin` and registered to :data:`~vllm.multimodal.MULTIMODAL_REGISTRY`.
For vLLM to recognize a new modality type, you have to create a new plugin and then pass it to :meth:`~vllm.multimodal.MultiModalRegistry.register_plugin`.
The remainder of this document details how to define custom :class:`~vllm.multimodal.MultiModalPlugin` s.
.. note::
This article is a work in progress.
..
TODO: Add more instructions on how to add new plugins once embeddings is in.

59
docs/source/design/multimodal/multimodal_index.rst → docs/source/design/multimodal/multimodal_index.md
View File

@ -1,66 +1,83 @@
.. _multi_modality:
(multi-modality)=
Multi-Modality
==============
# Multi-Modality
```{eval-rst}
.. currentmodule:: vllm.multimodal
```
vLLM provides experimental support for multi-modal models through the :mod:`vllm.multimodal` package.
vLLM provides experimental support for multi-modal models through the {mod}`vllm.multimodal` package.
Multi-modal inputs can be passed alongside text and token prompts to :ref:`supported models <supported_mm_models>`
via the ``multi_modal_data`` field in :class:`vllm.inputs.PromptType`.
Multi-modal inputs can be passed alongside text and token prompts to [supported models](#supported-mm-models)
via the `multi_modal_data` field in {class}`vllm.inputs.PromptType`.
Currently, vLLM only has built-in support for image data. You can extend vLLM to process additional modalities
by following :ref:`this guide <adding_multimodal_plugin>`.
by following [this guide](#adding-multimodal-plugin).
Looking to add your own multi-modal model? Please follow the instructions listed :ref:`here <enabling_multimodal_inputs>`.
Looking to add your own multi-modal model? Please follow the instructions listed [here](#enabling-multimodal-inputs).
Guides
++++++
## Guides
.. toctree::
:maxdepth: 1
```{toctree}
:maxdepth: 1
adding_multimodal_plugin
adding_multimodal_plugin
```
Module Contents
+++++++++++++++
## Module Contents
```{eval-rst}
.. automodule:: vllm.multimodal
```
Registry
--------
### Registry
```{eval-rst}
.. autodata:: vllm.multimodal.MULTIMODAL_REGISTRY
```
```{eval-rst}
.. autoclass:: vllm.multimodal.MultiModalRegistry
:members:
:show-inheritance:
```
Base Classes
------------
### Base Classes
```{eval-rst}
.. autodata:: vllm.multimodal.NestedTensors
```
```{eval-rst}
.. autodata:: vllm.multimodal.BatchedTensorInputs
```
```{eval-rst}
.. autoclass:: vllm.multimodal.MultiModalDataBuiltins
:members:
:show-inheritance:
```
```{eval-rst}
.. autodata:: vllm.multimodal.MultiModalDataDict
```
```{eval-rst}
.. autoclass:: vllm.multimodal.MultiModalKwargs
:members:
:show-inheritance:
```
```{eval-rst}
.. autoclass:: vllm.multimodal.MultiModalPlugin
:members:
:show-inheritance:
```
Image Classes
-------------
### Image Classes
```{eval-rst}
.. automodule:: vllm.multimodal.image
:members:
:show-inheritance:
```

54
docs/source/design/plugin_system.md Normal file
View File

@ -0,0 +1,54 @@
(plugin-system)=
# vLLM's Plugin System
The community frequently requests the ability to extend vLLM with custom features. To facilitate this, vLLM includes a plugin system that allows users to add custom features without modifying the vLLM codebase. This document explains how plugins work in vLLM and how to create a plugin for vLLM.
## How Plugins Work in vLLM
Plugins are user-registered code that vLLM executes. Given vLLM's architecture (see [](#arch-overview)), multiple processes may be involved, especially when using distributed inference with various parallelism techniques. To enable plugins successfully, every process created by vLLM needs to load the plugin. This is done by the [load_general_plugins](https://github.com/vllm-project/vllm/blob/c76ac49d266e27aa3fea84ef2df1f813d24c91c7/vllm/plugins/__init__.py#L16) function in the `vllm.plugins` module. This function is called for every process created by vLLM before it starts any work.
## How vLLM Discovers Plugins
vLLM's plugin system uses the standard Python `entry_points` mechanism. This mechanism allows developers to register functions in their Python packages for use by other packages. An example of a plugin:
```python
# inside `setup.py` file
from setuptools import setup
setup(name='vllm_add_dummy_model',
version='0.1',
packages=['vllm_add_dummy_model'],
entry_points={
'vllm.general_plugins':
["register_dummy_model = vllm_add_dummy_model:register"]
})
# inside `vllm_add_dummy_model.py` file
def register():
from vllm import ModelRegistry
if "MyLlava" not in ModelRegistry.get_supported_archs():
ModelRegistry.register_model("MyLlava",
"vllm_add_dummy_model.my_llava:MyLlava")
```
For more information on adding entry points to your package, please check the [official documentation](https://setuptools.pypa.io/en/latest/userguide/entry_point.html).
Every plugin has three parts:
1. **Plugin group**: The name of the entry point group. vLLM uses the entry point group `vllm.general_plugins` to register general plugins. This is the key of `entry_points` in the `setup.py` file. Always use `vllm.general_plugins` for vLLM's general plugins.
2. **Plugin name**: The name of the plugin. This is the value in the dictionary of the `entry_points` dictionary. In the example above, the plugin name is `register_dummy_model`. Plugins can be filtered by their names using the `VLLM_PLUGINS` environment variable. To load only a specific plugin, set `VLLM_PLUGINS` to the plugin name.
3. **Plugin value**: The fully qualified name of the function to register in the plugin system. In the example above, the plugin value is `vllm_add_dummy_model:register`, which refers to a function named `register` in the `vllm_add_dummy_model` module.
## What Can Plugins Do?
Currently, the primary use case for plugins is to register custom, out-of-the-tree models into vLLM. This is done by calling `ModelRegistry.register_model` to register the model. In the future, the plugin system may be extended to support more features, such as swapping in custom implementations for certain classes in vLLM.
## Guidelines for Writing Plugins
- **Being re-entrant**: The function specified in the entry point should be re-entrant, meaning it can be called multiple times without causing issues. This is necessary because the function might be called multiple times in some processes.
## Compatibility Guarantee
vLLM guarantees the interface of documented plugins, such as `ModelRegistry.register_model`, will always be available for plugins to register models. However, it is the responsibility of plugin developers to ensure their plugins are compatible with the version of vLLM they are targeting. For example, `"vllm_add_dummy_model.my_llava:MyLlava"` should be compatible with the version of vLLM that the plugin targets. The interface for the model may change during vLLM's development.

62
docs/source/design/plugin_system.rst
View File

@ -1,62 +0,0 @@
.. _plugin_system:
vLLM's Plugin System
====================
The community frequently requests the ability to extend vLLM with custom features. To facilitate this, vLLM includes a plugin system that allows users to add custom features without modifying the vLLM codebase. This document explains how plugins work in vLLM and how to create a plugin for vLLM.
How Plugins Work in vLLM
------------------------
Plugins are user-registered code that vLLM executes. Given vLLM's architecture (see :ref:`arch_overview`), multiple processes may be involved, especially when using distributed inference with various parallelism techniques. To enable plugins successfully, every process created by vLLM needs to load the plugin. This is done by the `load_general_plugins <https://github.com/vllm-project/vllm/blob/c76ac49d266e27aa3fea84ef2df1f813d24c91c7/vllm/plugins/__init__.py#L16>`__ function in the ``vllm.plugins`` module. This function is called for every process created by vLLM before it starts any work.
How vLLM Discovers Plugins
--------------------------
vLLM's plugin system uses the standard Python ``entry_points`` mechanism. This mechanism allows developers to register functions in their Python packages for use by other packages. An example of a plugin:
.. code-block:: python
# inside `setup.py` file
from setuptools import setup
setup(name='vllm_add_dummy_model',
version='0.1',
packages=['vllm_add_dummy_model'],
entry_points={
'vllm.general_plugins':
["register_dummy_model = vllm_add_dummy_model:register"]
})
# inside `vllm_add_dummy_model.py` file
def register():
from vllm import ModelRegistry
if "MyLlava" not in ModelRegistry.get_supported_archs():
ModelRegistry.register_model("MyLlava",
"vllm_add_dummy_model.my_llava:MyLlava")
For more information on adding entry points to your package, please check the `official documentation <https://setuptools.pypa.io/en/latest/userguide/entry_point.html>`__.
Every plugin has three parts:
1. **Plugin group**: The name of the entry point group. vLLM uses the entry point group ``vllm.general_plugins`` to register general plugins. This is the key of ``entry_points`` in the ``setup.py`` file. Always use ``vllm.general_plugins`` for vLLM's general plugins.
2. **Plugin name**: The name of the plugin. This is the value in the dictionary of the ``entry_points`` dictionary. In the example above, the plugin name is ``register_dummy_model``. Plugins can be filtered by their names using the ``VLLM_PLUGINS`` environment variable. To load only a specific plugin, set ``VLLM_PLUGINS`` to the plugin name.
3. **Plugin value**: The fully qualified name of the function to register in the plugin system. In the example above, the plugin value is ``vllm_add_dummy_model:register``, which refers to a function named ``register`` in the ``vllm_add_dummy_model`` module.
What Can Plugins Do?
--------------------
Currently, the primary use case for plugins is to register custom, out-of-the-tree models into vLLM. This is done by calling ``ModelRegistry.register_model`` to register the model. In the future, the plugin system may be extended to support more features, such as swapping in custom implementations for certain classes in vLLM.
Guidelines for Writing Plugins
------------------------------
- **Being re-entrant**: The function specified in the entry point should be re-entrant, meaning it can be called multiple times without causing issues. This is necessary because the function might be called multiple times in some processes.
Compatibility Guarantee
-----------------------
vLLM guarantees the interface of documented plugins, such as ``ModelRegistry.register_model``, will always be available for plugins to register models. However, it is the responsibility of plugin developers to ensure their plugins are compatible with the version of vLLM they are targeting. For example, ``"vllm_add_dummy_model.my_llava:MyLlava"`` should be compatible with the version of vLLM that the plugin targets. The interface for the model may change during vLLM's development.

5
docs/source/dev/engine/async_llm_engine.rst → docs/source/dev/engine/async_llm_engine.md
View File

@ -1,6 +1,7 @@
AsyncLLMEngine
=================================
# AsyncLLMEngine
```{eval-rst}
.. autoclass:: vllm.AsyncLLMEngine
:members:
:show-inheritance:
```

17
docs/source/dev/engine/engine_index.md Normal file
View File

@ -0,0 +1,17 @@
# vLLM Engine
```{eval-rst}
.. automodule:: vllm.engine
```
```{eval-rst}
.. currentmodule:: vllm.engine
```
```{toctree}
:caption: Engines
:maxdepth: 2
llm_engine
async_llm_engine
```

13
docs/source/dev/engine/engine_index.rst
View File

@ -1,13 +0,0 @@
vLLM Engine
=================================
.. automodule:: vllm.engine
.. currentmodule:: vllm.engine
.. toctree::
:maxdepth: 2
:caption: Engines
llm_engine
async_llm_engine

5
docs/source/dev/engine/llm_engine.rst → docs/source/dev/engine/llm_engine.md
View File

@ -1,6 +1,7 @@
LLMEngine
=================================
# LLMEngine
```{eval-rst}
.. autoclass:: vllm.LLMEngine
:members:
:show-inheritance:
```

5
docs/source/dev/offline_inference/llm.rst → docs/source/dev/offline_inference/llm.md
View File

@ -1,6 +1,7 @@
LLM Class
=========
# LLM Class
```{eval-rst}
.. autoclass:: vllm.LLM
:members:
:show-inheritance:
```

9
docs/source/dev/offline_inference/llm_inputs.rst → docs/source/dev/offline_inference/llm_inputs.md
View File

@ -1,14 +1,19 @@
LLM Inputs
==========
# LLM Inputs
```{eval-rst}
.. autodata:: vllm.inputs.PromptType
```
```{eval-rst}
.. autoclass:: vllm.inputs.TextPrompt
:show-inheritance:
:members:
:member-order: bysource
```
```{eval-rst}
.. autoclass:: vllm.inputs.TokensPrompt
:show-inheritance:
:members:
:member-order: bysource
```

8
docs/source/dev/offline_inference/offline_index.md Normal file
View File

@ -0,0 +1,8 @@
# Offline Inference
```{toctree}
:maxdepth: 1
llm
llm_inputs
```

8
docs/source/dev/offline_inference/offline_index.rst
View File

@ -1,8 +0,0 @@
Offline Inference
=================================
.. toctree::
:maxdepth: 1
llm
llm_inputs

5
docs/source/dev/pooling_params.rst → docs/source/dev/pooling_params.md
View File

@ -1,5 +1,6 @@
Pooling Parameters
==================
# Pooling Parameters
```{eval-rst}
.. autoclass:: vllm.PoolingParams
:members:
```

5
docs/source/dev/sampling_params.rst → docs/source/dev/sampling_params.md
View File

@ -1,5 +1,6 @@
Sampling Parameters
===================
# Sampling Parameters
```{eval-rst}
.. autoclass:: vllm.SamplingParams
:members:
```

24
docs/source/generate_examples.py
View File

@ -15,18 +15,12 @@ def fix_case(text: str) -> str:
return text
def underline(title: str, character: str = "=") -> str:
return f"{title}\n{character * len(title)}"
def generate_title(filename: str) -> str:
# Turn filename into a title
title = filename.replace("_", " ").title()
# Handle acronyms and names
title = fix_case(title)
# Underline title
title = underline(title)
return title
return f"# {title}"
def generate_examples():
@ -38,7 +32,7 @@ def generate_examples():
# Destination paths
doc_dir = root_dir / "docs/source/getting_started/examples"
doc_paths = [doc_dir / f"{path.stem}.rst" for path in script_paths]
doc_paths = [doc_dir / f"{path.stem}.md" for path in script_paths]
# Generate the example docs for each example script
for script_path, doc_path in zip(script_paths, doc_paths):
@ -46,16 +40,16 @@ def generate_examples():
# Make script_path relative to doc_path and call it include_path
include_path = '../../../..' / script_path.relative_to(root_dir)
content = (f"{generate_title(doc_path.stem)}\n\n"
f"Source {script_url}.\n\n"
f".. literalinclude:: {include_path}\n"
" :language: python\n"
" :linenos:\n")
f"Source: <{script_url}>.\n\n"
f"```{{literalinclude}} {include_path}\n"
":language: python\n"
":linenos:\n```")
with open(doc_path, "w+") as f:
f.write(content)
# Generate the toctree for the example scripts
with open(doc_dir / "examples_index.template.rst") as f:
with open(doc_dir / "examples_index.template.md") as f:
examples_index = f.read()
with open(doc_dir / "examples_index.rst", "w+") as f:
example_docs = "\n ".join(path.stem for path in script_paths)
with open(doc_dir / "examples_index.md", "w+") as f:
example_docs = "\n".join(path.stem + ".md" for path in script_paths)
f.write(examples_index.replace(r"%EXAMPLE_DOCS%", example_docs))

163
docs/source/getting_started/amd-installation.md Normal file
View File

@ -0,0 +1,163 @@
(installation-rocm)=
# Installation with ROCm
vLLM supports AMD GPUs with ROCm 6.2.
## Requirements
- OS: Linux
- Python: 3.9 -- 3.12
- GPU: MI200s (gfx90a), MI300 (gfx942), Radeon RX 7900 series (gfx1100)
- ROCm 6.2
Installation options:
1. [Build from source with docker](#build-from-source-docker-rocm)
2. [Build from source](#build-from-source-rocm)
(build-from-source-docker-rocm)=
## Option 1: Build from source with docker (recommended)
You can build and install vLLM from source.
First, build a docker image from [Dockerfile.rocm](https://github.com/vllm-project/vllm/blob/main/Dockerfile.rocm) and launch a docker container from the image.
It is important that the user kicks off the docker build using buildkit. Either the user put DOCKER_BUILDKIT=1 as environment variable when calling docker build command, or the user needs to setup buildkit in the docker daemon configuration /etc/docker/daemon.json as follows and restart the daemon:
```console
{
"features": {
"buildkit": true
}
}
```
[Dockerfile.rocm](https://github.com/vllm-project/vllm/blob/main/Dockerfile.rocm) uses ROCm 6.2 by default, but also supports ROCm 5.7, 6.0 and 6.1 in older vLLM branches.
It provides flexibility to customize the build of docker image using the following arguments:
- `BASE_IMAGE`: specifies the base image used when running `docker build`, specifically the PyTorch on ROCm base image.
- `BUILD_FA`: specifies whether to build CK flash-attention. The default is 1. For [Radeon RX 7900 series (gfx1100)](https://rocm.docs.amd.com/projects/radeon/en/latest/index.html), this should be set to 0 before flash-attention supports this target.
- `FX_GFX_ARCHS`: specifies the GFX architecture that is used to build CK flash-attention, for example, `gfx90a;gfx942` for MI200 and MI300. The default is `gfx90a;gfx942`
- `FA_BRANCH`: specifies the branch used to build the CK flash-attention in [ROCm's flash-attention repo](https://github.com/ROCmSoftwarePlatform/flash-attention). The default is `ae7928c`
- `BUILD_TRITON`: specifies whether to build triton flash-attention. The default value is 1.
Their values can be passed in when running `docker build` with `--build-arg` options.
To build vllm on ROCm 6.2 for MI200 and MI300 series, you can use the default:
```console
$ DOCKER_BUILDKIT=1 docker build -f Dockerfile.rocm -t vllm-rocm .
```
To build vllm on ROCm 6.2 for Radeon RX7900 series (gfx1100), you should specify `BUILD_FA` as below:
```console
$ DOCKER_BUILDKIT=1 docker build --build-arg BUILD_FA="0" -f Dockerfile.rocm -t vllm-rocm .
```
To run the above docker image `vllm-rocm`, use the below command:
```console
$ docker run -it \
--network=host \
--group-add=video \
--ipc=host \
--cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined \
--device /dev/kfd \
--device /dev/dri \
-v <path/to/model>:/app/model \
vllm-rocm \
bash
```
Where the `<path/to/model>` is the location where the model is stored, for example, the weights for llama2 or llama3 models.
(build-from-source-rocm)=
## Option 2: Build from source
0. Install prerequisites (skip if you are already in an environment/docker with the following installed):
- [ROCm](https://rocm.docs.amd.com/en/latest/deploy/linux/index.html)
- [PyTorch](https://pytorch.org/)
For installing PyTorch, you can start from a fresh docker image, e.g, `rocm/pytorch:rocm6.2_ubuntu20.04_py3.9_pytorch_release_2.3.0`, `rocm/pytorch-nightly`.
Alternatively, you can install PyTorch using PyTorch wheels. You can check PyTorch installation guide in PyTorch [Getting Started](https://pytorch.org/get-started/locally/)
1. Install [Triton flash attention for ROCm](https://github.com/ROCm/triton)
Install ROCm's Triton flash attention (the default triton-mlir branch) following the instructions from [ROCm/triton](https://github.com/ROCm/triton/blob/triton-mlir/README.md)
```console
$ python3 -m pip install ninja cmake wheel pybind11
$ pip uninstall -y triton
$ git clone https://github.com/OpenAI/triton.git
$ cd triton
$ git checkout e192dba
$ cd python
$ pip3 install .
$ cd ../..
```
```{note}
- If you see HTTP issue related to downloading packages during building triton, please try again as the HTTP error is intermittent.
```
2. Optionally, if you choose to use CK flash attention, you can install [flash attention for ROCm](https://github.com/ROCm/flash-attention/tree/ck_tile)
Install ROCm's flash attention (v2.5.9.post1) following the instructions from [ROCm/flash-attention](https://github.com/ROCm/flash-attention/tree/ck_tile#amd-gpurocm-support)
Alternatively, wheels intended for vLLM use can be accessed under the releases.
For example, for ROCm 6.2, suppose your gfx arch is `gfx90a`. To get your gfx architecture, run `rocminfo |grep gfx`.
```console
$ git clone https://github.com/ROCm/flash-attention.git
$ cd flash-attention
$ git checkout 3cea2fb
$ git submodule update --init
$ GPU_ARCHS="gfx90a" python3 setup.py install
$ cd ..
```
```{note}
- You might need to downgrade the "ninja" version to 1.10 it is not used when compiling flash-attention-2 (e.g. `pip install ninja==1.10.2.4`)
```
3. Build vLLM. For example, vLLM on ROCM 6.2 can be built with the following steps:
```bash
$ pip install --upgrade pip
# Install PyTorch
$ pip uninstall torch -y
$ pip install --no-cache-dir --pre torch==2.6.0.dev20241024 --index-url https://download.pytorch.org/whl/nightly/rocm6.2
# Build & install AMD SMI
$ pip install /opt/rocm/share/amd_smi
# Install dependencies
$ pip install --upgrade numba scipy huggingface-hub[cli]
$ pip install "numpy<2"
$ pip install -r requirements-rocm.txt
# Build vLLM for MI210/MI250/MI300.
$ export PYTORCH_ROCM_ARCH="gfx90a;gfx942"
$ python3 setup.py develop
```
This may take 5-10 minutes. Currently, {code}`pip install .` does not work for ROCm installation.
```{tip}
- Triton flash attention is used by default. For benchmarking purposes, it is recommended to run a warm up step before collecting perf numbers.
- Triton flash attention does not currently support sliding window attention. If using half precision, please use CK flash-attention for sliding window support.
- To use CK flash-attention or PyTorch naive attention, please use this flag `export VLLM_USE_TRITON_FLASH_ATTN=0` to turn off triton flash attention.
- The ROCm version of PyTorch, ideally, should match the ROCm driver version.
```
```{tip}
- For MI300x (gfx942) users, to achieve optimal performance, please refer to [MI300x tuning guide](https://rocm.docs.amd.com/en/latest/how-to/tuning-guides/mi300x/index.html) for performance optimization and tuning tips on system and workflow level.
For vLLM, please refer to [vLLM performance optimization](https://rocm.docs.amd.com/en/latest/how-to/tuning-guides/mi300x/workload.html#vllm-performance-optimization).
```

178
docs/source/getting_started/amd-installation.rst
View File

@ -1,178 +0,0 @@
.. _installation_rocm:
Installation with ROCm
======================
vLLM supports AMD GPUs with ROCm 6.2.
Requirements
------------
* OS: Linux
* Python: 3.9 -- 3.12
* GPU: MI200s (gfx90a), MI300 (gfx942), Radeon RX 7900 series (gfx1100)
* ROCm 6.2
Installation options:
#. :ref:`Build from source with docker <build_from_source_docker_rocm>`
#. :ref:`Build from source <build_from_source_rocm>`
.. _build_from_source_docker_rocm:
Option 1: Build from source with docker (recommended)
-----------------------------------------------------
You can build and install vLLM from source.
First, build a docker image from `Dockerfile.rocm <https://github.com/vllm-project/vllm/blob/main/Dockerfile.rocm>`_ and launch a docker container from the image.
It is important that the user kicks off the docker build using buildkit. Either the user put DOCKER_BUILDKIT=1 as environment variable when calling docker build command, or the user needs to setup buildkit in the docker daemon configuration /etc/docker/daemon.json as follows and restart the daemon:
.. code-block:: console
{
"features": {
"buildkit": true
}
}
`Dockerfile.rocm <https://github.com/vllm-project/vllm/blob/main/Dockerfile.rocm>`_ uses ROCm 6.2 by default, but also supports ROCm 5.7, 6.0 and 6.1 in older vLLM branches.
It provides flexibility to customize the build of docker image using the following arguments:
* `BASE_IMAGE`: specifies the base image used when running ``docker build``, specifically the PyTorch on ROCm base image.
* `BUILD_FA`: specifies whether to build CK flash-attention. The default is 1. For `Radeon RX 7900 series (gfx1100) <https://rocm.docs.amd.com/projects/radeon/en/latest/index.html>`_, this should be set to 0 before flash-attention supports this target.
* `FX_GFX_ARCHS`: specifies the GFX architecture that is used to build CK flash-attention, for example, `gfx90a;gfx942` for MI200 and MI300. The default is `gfx90a;gfx942`
* `FA_BRANCH`: specifies the branch used to build the CK flash-attention in `ROCm's flash-attention repo <https://github.com/ROCmSoftwarePlatform/flash-attention>`_. The default is `ae7928c`
* `BUILD_TRITON`: specifies whether to build triton flash-attention. The default value is 1.
Their values can be passed in when running ``docker build`` with ``--build-arg`` options.
To build vllm on ROCm 6.2 for MI200 and MI300 series, you can use the default:
.. code-block:: console
$ DOCKER_BUILDKIT=1 docker build -f Dockerfile.rocm -t vllm-rocm .
To build vllm on ROCm 6.2 for Radeon RX7900 series (gfx1100), you should specify ``BUILD_FA`` as below:
.. code-block:: console
$ DOCKER_BUILDKIT=1 docker build --build-arg BUILD_FA="0" -f Dockerfile.rocm -t vllm-rocm .
To run the above docker image ``vllm-rocm``, use the below command:
.. code-block:: console
$ docker run -it \
--network=host \
--group-add=video \
--ipc=host \
--cap-add=SYS_PTRACE \
--security-opt seccomp=unconfined \
--device /dev/kfd \
--device /dev/dri \
-v <path/to/model>:/app/model \
vllm-rocm \
bash
Where the `<path/to/model>` is the location where the model is stored, for example, the weights for llama2 or llama3 models.
.. _build_from_source_rocm:
Option 2: Build from source
---------------------------
0. Install prerequisites (skip if you are already in an environment/docker with the following installed):
- `ROCm <https://rocm.docs.amd.com/en/latest/deploy/linux/index.html>`_
- `PyTorch <https://pytorch.org/>`_
For installing PyTorch, you can start from a fresh docker image, e.g, `rocm/pytorch:rocm6.2_ubuntu20.04_py3.9_pytorch_release_2.3.0`, `rocm/pytorch-nightly`.
Alternatively, you can install PyTorch using PyTorch wheels. You can check PyTorch installation guide in PyTorch `Getting Started <https://pytorch.org/get-started/locally/>`_
1. Install `Triton flash attention for ROCm <https://github.com/ROCm/triton>`_
Install ROCm's Triton flash attention (the default triton-mlir branch) following the instructions from `ROCm/triton <https://github.com/ROCm/triton/blob/triton-mlir/README.md>`_
.. code-block:: console
$ python3 -m pip install ninja cmake wheel pybind11
$ pip uninstall -y triton
$ git clone https://github.com/OpenAI/triton.git
$ cd triton
$ git checkout e192dba
$ cd python
$ pip3 install .
$ cd ../..
.. note::
- If you see HTTP issue related to downloading packages during building triton, please try again as the HTTP error is intermittent.
2. Optionally, if you choose to use CK flash attention, you can install `flash attention for ROCm <https://github.com/ROCm/flash-attention/tree/ck_tile>`_
Install ROCm's flash attention (v2.5.9.post1) following the instructions from `ROCm/flash-attention <https://github.com/ROCm/flash-attention/tree/ck_tile#amd-gpurocm-support>`_
Alternatively, wheels intended for vLLM use can be accessed under the releases.
For example, for ROCm 6.2, suppose your gfx arch is `gfx90a`.
Note to get your gfx architecture, run `rocminfo |grep gfx`.
.. code-block:: console
$ git clone https://github.com/ROCm/flash-attention.git
$ cd flash-attention
$ git checkout 3cea2fb
$ git submodule update --init
$ GPU_ARCHS="gfx90a" python3 setup.py install
$ cd ..
.. note::
- You might need to downgrade the "ninja" version to 1.10 it is not used when compiling flash-attention-2 (e.g. `pip install ninja==1.10.2.4`)
3. Build vLLM.
For example, vLLM on ROCM 6.2 can be built with the following steps:
.. code-block:: console
$ pip install --upgrade pip
$ # Install PyTorch
$ pip uninstall torch -y
$ pip install --no-cache-dir --pre torch==2.6.0.dev20241024 --index-url https://download.pytorch.org/whl/nightly/rocm6.2
$ # Build & install AMD SMI
$ pip install /opt/rocm/share/amd_smi
$ # Install dependencies
$ pip install --upgrade numba scipy huggingface-hub[cli]
$ pip install "numpy<2"
$ pip install -r requirements-rocm.txt
$ # Build vLLM for MI210/MI250/MI300.
$ export PYTORCH_ROCM_ARCH="gfx90a;gfx942"
$ python3 setup.py develop
This may take 5-10 minutes. Currently, :code:`pip install .` does not work for ROCm installation.
.. tip::
- Triton flash attention is used by default. For benchmarking purposes, it is recommended to run a warm up step before collecting perf numbers.
- Triton flash attention does not currently support sliding window attention. If using half precision, please use CK flash-attention for sliding window support.
- To use CK flash-attention or PyTorch naive attention, please use this flag ``export VLLM_USE_TRITON_FLASH_ATTN=0`` to turn off triton flash attention.
- The ROCm version of PyTorch, ideally, should match the ROCm driver version.
.. tip::
- For MI300x (gfx942) users, to achieve optimal performance, please refer to `MI300x tuning guide <https://rocm.docs.amd.com/en/latest/how-to/tuning-guides/mi300x/index.html>`_ for performance optimization and tuning tips on system and workflow level.
For vLLM, please refer to `vLLM performance optimization <https://rocm.docs.amd.com/en/latest/how-to/tuning-guides/mi300x/workload.html#vllm-performance-optimization>`_.

46
docs/source/getting_started/arm-installation.md Normal file
View File

@ -0,0 +1,46 @@
(installation-arm)=
# Installation for ARM CPUs
vLLM has been adapted to work on ARM64 CPUs with NEON support, leveraging the CPU backend initially developed for the x86 platform. This guide provides installation instructions specific to ARM. For additional details on supported features, refer to the x86 platform documentation covering:
- CPU backend inference capabilities
- Relevant runtime environment variables
- Performance optimization tips
ARM CPU backend currently supports Float32, FP16 and BFloat16 datatypes.
Contents:
1. [Requirements](#arm-backend-requirements)
2. [Quick Start with Dockerfile](#arm-backend-quick-start-dockerfile)
3. [Building from Source](#build-arm-backend-from-source)
(arm-backend-requirements)=
## Requirements
- **Operating System**: Linux or macOS
- **Compiler**: gcc/g++ >= 12.3.0 (optional, but recommended)
- **Instruction Set Architecture (ISA)**: NEON support is required
(arm-backend-quick-start-dockerfile)=
## Quick Start with Dockerfile
You can quickly set up vLLM on ARM using Docker:
```console
$ docker build -f Dockerfile.arm -t vllm-cpu-env --shm-size=4g .
$ docker run -it \
--rm \
--network=host \
--cpuset-cpus=<cpu-id-list, optional> \
--cpuset-mems=<memory-node, optional> \
vllm-cpu-env
```
(build-arm-backend-from-source)=
## Building from Source
To build vLLM from source on Ubuntu 22.04 or other Linux distributions, follow a similar process as with x86. Testing has been conducted on AWS Graviton3 instances for compatibility.

50
docs/source/getting_started/arm-installation.rst
View File

@ -1,50 +0,0 @@
.. _installation_arm:
Installation for ARM CPUs
=========================
vLLM has been adapted to work on ARM64 CPUs with NEON support, leveraging the CPU backend initially developed for the x86 platform. This guide provides installation instructions specific to ARM. For additional details on supported features, refer to the x86 platform documentation covering:
* CPU backend inference capabilities
* Relevant runtime environment variables
* Performance optimization tips
ARM CPU backend currently supports Float32, FP16 and BFloat16 datatypes.
Contents:
1. :ref:`Requirements <arm_backend_requirements>`
2. :ref:`Quick Start with Dockerfile <arm_backend_quick_start_dockerfile>`
3. :ref:`Building from Source <build_arm_backend_from_source>`
.. _arm_backend_requirements:
Requirements
------------
* **Operating System**: Linux or macOS
* **Compiler**: gcc/g++ >= 12.3.0 (optional, but recommended)
* **Instruction Set Architecture (ISA)**: NEON support is required
.. _arm_backend_quick_start_dockerfile:
Quick Start with Dockerfile
---------------------------
You can quickly set up vLLM on ARM using Docker:
.. code-block:: console
$ docker build -f Dockerfile.arm -t vllm-cpu-env --shm-size=4g .
$ docker run -it \
--rm \
--network=host \
--cpuset-cpus=<cpu-id-list, optional> \
--cpuset-mems=<memory-node, optional> \
vllm-cpu-env
.. _build_arm_backend_from_source:
Building from Source
--------------------
To build vLLM from source on Ubuntu 22.04 or other Linux distributions, follow a similar process as with x86. Testing has been conducted on AWS Graviton3 instances for compatibility.

154
docs/source/getting_started/cpu-installation.md Normal file
View File

@ -0,0 +1,154 @@
(installation-cpu)=
# Installation with CPU
vLLM initially supports basic model inferencing and serving on x86 CPU platform, with data types FP32, FP16 and BF16. vLLM CPU backend supports the following vLLM features:
- Tensor Parallel
- Model Quantization (`INT8 W8A8, AWQ`)
- Chunked-prefill
- Prefix-caching
- FP8-E5M2 KV-Caching (TODO)
Table of contents:
1. [Requirements](#cpu-backend-requirements)
2. [Quick start using Dockerfile](#cpu-backend-quick-start-dockerfile)
3. [Build from source](#build-cpu-backend-from-source)
4. [Related runtime environment variables](#env-intro)
5. [Intel Extension for PyTorch](#ipex-guidance)
6. [Performance tips](#cpu-backend-performance-tips)
(cpu-backend-requirements)=
## Requirements
- OS: Linux
- Compiler: gcc/g++>=12.3.0 (optional, recommended)
- Instruction set architecture (ISA) requirement: AVX512 (optional, recommended)
(cpu-backend-quick-start-dockerfile)=
## Quick start using Dockerfile
```console
$ docker build -f Dockerfile.cpu -t vllm-cpu-env --shm-size=4g .
$ docker run -it \
--rm \
--network=host \
--cpuset-cpus=<cpu-id-list, optional> \
--cpuset-mems=<memory-node, optional> \
vllm-cpu-env
```
(build-cpu-backend-from-source)=
## Build from source
- First, install recommended compiler. We recommend to use `gcc/g++ >= 12.3.0` as the default compiler to avoid potential problems. For example, on Ubuntu 22.4, you can run:
```console
$ sudo apt-get update -y
$ sudo apt-get install -y gcc-12 g++-12 libnuma-dev
$ sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 10 --slave /usr/bin/g++ g++ /usr/bin/g++-12
```
- Second, install Python packages for vLLM CPU backend building:
```console
$ pip install --upgrade pip
$ pip install cmake>=3.26 wheel packaging ninja "setuptools-scm>=8" numpy
$ pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
```
- Finally, build and install vLLM CPU backend:
```console
$ VLLM_TARGET_DEVICE=cpu python setup.py install
```
```{note}
- AVX512_BF16 is an extension ISA provides native BF16 data type conversion and vector product instructions, will brings some performance improvement compared with pure AVX512. The CPU backend build script will check the host CPU flags to determine whether to enable AVX512_BF16.
- If you want to force enable AVX512_BF16 for the cross-compilation, please set environment variable VLLM_CPU_AVX512BF16=1 before the building.
```
(env-intro)=
## Related runtime environment variables
- `VLLM_CPU_KVCACHE_SPACE`: specify the KV Cache size (e.g, `VLLM_CPU_KVCACHE_SPACE=40` means 40 GB space for KV cache), larger setting will allow vLLM running more requests in parallel. This parameter should be set based on the hardware configuration and memory management pattern of users.
- `VLLM_CPU_OMP_THREADS_BIND`: specify the CPU cores dedicated to the OpenMP threads. For example, `VLLM_CPU_OMP_THREADS_BIND=0-31` means there will be 32 OpenMP threads bound on 0-31 CPU cores. `VLLM_CPU_OMP_THREADS_BIND=0-31|32-63` means there will be 2 tensor parallel processes, 32 OpenMP threads of rank0 are bound on 0-31 CPU cores, and the OpenMP threads of rank1 are bound on 32-63 CPU cores.
(ipex-guidance)=
## Intel Extension for PyTorch
- [Intel Extension for PyTorch (IPEX)](https://github.com/intel/intel-extension-for-pytorch) extends PyTorch with up-to-date features optimizations for an extra performance boost on Intel hardware.
(cpu-backend-performance-tips)=
## Performance tips
- We highly recommend to use TCMalloc for high performance memory allocation and better cache locality. For example, on Ubuntu 22.4, you can run:
```console
$ sudo apt-get install libtcmalloc-minimal4 # install TCMalloc library
$ find / -name *libtcmalloc* # find the dynamic link library path
$ export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4:$LD_PRELOAD # prepend the library to LD_PRELOAD
$ python examples/offline_inference.py # run vLLM
```
- When using the online serving, it is recommended to reserve 1-2 CPU cores for the serving framework to avoid CPU oversubscription. For example, on a platform with 32 physical CPU cores, reserving CPU 30 and 31 for the framework and using CPU 0-29 for OpenMP:
```console
$ export VLLM_CPU_KVCACHE_SPACE=40
$ export VLLM_CPU_OMP_THREADS_BIND=0-29
$ vllm serve facebook/opt-125m
```
- If using vLLM CPU backend on a machine with hyper-threading, it is recommended to bind only one OpenMP thread on each physical CPU core using `VLLM_CPU_OMP_THREADS_BIND`. On a hyper-threading enabled platform with 16 logical CPU cores / 8 physical CPU cores:
```console
$ lscpu -e # check the mapping between logical CPU cores and physical CPU cores
# The "CPU" column means the logical CPU core IDs, and the "CORE" column means the physical core IDs. On this platform, two logical cores are sharing one physical core.
CPU NODE SOCKET CORE L1d:L1i:L2:L3 ONLINE MAXMHZ MINMHZ MHZ
0 0 0 0 0:0:0:0 yes 2401.0000 800.0000 800.000
1 0 0 1 1:1:1:0 yes 2401.0000 800.0000 800.000
2 0 0 2 2:2:2:0 yes 2401.0000 800.0000 800.000
3 0 0 3 3:3:3:0 yes 2401.0000 800.0000 800.000
4 0 0 4 4:4:4:0 yes 2401.0000 800.0000 800.000
5 0 0 5 5:5:5:0 yes 2401.0000 800.0000 800.000
6 0 0 6 6:6:6:0 yes 2401.0000 800.0000 800.000
7 0 0 7 7:7:7:0 yes 2401.0000 800.0000 800.000
8 0 0 0 0:0:0:0 yes 2401.0000 800.0000 800.000
9 0 0 1 1:1:1:0 yes 2401.0000 800.0000 800.000
10 0 0 2 2:2:2:0 yes 2401.0000 800.0000 800.000
11 0 0 3 3:3:3:0 yes 2401.0000 800.0000 800.000
12 0 0 4 4:4:4:0 yes 2401.0000 800.0000 800.000
13 0 0 5 5:5:5:0 yes 2401.0000 800.0000 800.000
14 0 0 6 6:6:6:0 yes 2401.0000 800.0000 800.000
15 0 0 7 7:7:7:0 yes 2401.0000 800.0000 800.000
# On this platform, it is recommend to only bind openMP threads on logical CPU cores 0-7 or 8-15
$ export VLLM_CPU_OMP_THREADS_BIND=0-7
$ python examples/offline_inference.py
```
- If using vLLM CPU backend on a multi-socket machine with NUMA, be aware to set CPU cores using `VLLM_CPU_OMP_THREADS_BIND` to avoid cross NUMA node memory access.
## CPU Backend Considerations
- The CPU backend significantly differs from the GPU backend since the vLLM architecture was originally optimized for GPU use. A number of optimizations are needed to enhance its performance.
- Decouple the HTTP serving components from the inference components. In a GPU backend configuration, the HTTP serving and tokenization tasks operate on the CPU, while inference runs on the GPU, which typically does not pose a problem. However, in a CPU-based setup, the HTTP serving and tokenization can cause significant context switching and reduced cache efficiency. Therefore, it is strongly recommended to segregate these two components for improved performance.
- On CPU based setup with NUMA enabled, the memory access performance may be largely impacted by the [topology](https://github.com/intel/intel-extension-for-pytorch/blob/main/docs/tutorials/performance_tuning/tuning_guide.md#non-uniform-memory-access-numa). For NUMA architecture, two optimizations are to recommended: Tensor Parallel or Data Parallel.
- Using Tensor Parallel for a latency constraints deployment: following GPU backend design, a Megatron-LM's parallel algorithm will be used to shard the model, based on the number of NUMA nodes (e.g. TP = 2 for a two NUMA node system). With [TP feature on CPU](https://github.com/vllm-project/vllm/pull/6125) merged, Tensor Parallel is supported for serving and offline inferencing. In general each NUMA node is treated as one GPU card. Below is the example script to enable Tensor Parallel = 2 for serving:
```console
$ VLLM_CPU_KVCACHE_SPACE=40 VLLM_CPU_OMP_THREADS_BIND="0-31|32-63" vllm serve meta-llama/Llama-2-7b-chat-hf -tp=2 --distributed-executor-backend mp
```
- Using Data Parallel for maximum throughput: to launch an LLM serving endpoint on each NUMA node along with one additional load balancer to dispatch the requests to those endpoints. Common solutions like [Nginx](../serving/deploying_with_nginx) or HAProxy are recommended. Anyscale Ray project provides the feature on LLM [serving](https://docs.ray.io/en/latest/serve/index.html). Here is the example to setup a scalable LLM serving with [Ray Serve](https://github.com/intel/llm-on-ray/blob/main/docs/setup.md).

164
docs/source/getting_started/cpu-installation.rst
View File

@ -1,164 +0,0 @@
.. _installation_cpu:
Installation with CPU
========================
vLLM initially supports basic model inferencing and serving on x86 CPU platform, with data types FP32, FP16 and BF16. vLLM CPU backend supports the following vLLM features:
- Tensor Parallel
- Model Quantization (``INT8 W8A8, AWQ``)
- Chunked-prefill
- Prefix-caching
- FP8-E5M2 KV-Caching (TODO)
Table of contents:
#. :ref:`Requirements <cpu_backend_requirements>`
#. :ref:`Quick start using Dockerfile <cpu_backend_quick_start_dockerfile>`
#. :ref:`Build from source <build_cpu_backend_from_source>`
#. :ref:`Related runtime environment variables <env_intro>`
#. :ref:`Intel Extension for PyTorch <ipex_guidance>`
#. :ref:`Performance tips <cpu_backend_performance_tips>`
.. _cpu_backend_requirements:
Requirements
------------
* OS: Linux
* Compiler: gcc/g++>=12.3.0 (optional, recommended)
* Instruction set architecture (ISA) requirement: AVX512 (optional, recommended)
.. _cpu_backend_quick_start_dockerfile:
Quick start using Dockerfile
----------------------------
.. code-block:: console
$ docker build -f Dockerfile.cpu -t vllm-cpu-env --shm-size=4g .
$ docker run -it \
--rm \
--network=host \
--cpuset-cpus=<cpu-id-list, optional> \
--cpuset-mems=<memory-node, optional> \
vllm-cpu-env
.. _build_cpu_backend_from_source:
Build from source
-----------------
- First, install recommended compiler. We recommend to use ``gcc/g++ >= 12.3.0`` as the default compiler to avoid potential problems. For example, on Ubuntu 22.4, you can run:
.. code-block:: console
$ sudo apt-get update -y
$ sudo apt-get install -y gcc-12 g++-12 libnuma-dev
$ sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 10 --slave /usr/bin/g++ g++ /usr/bin/g++-12
- Second, install Python packages for vLLM CPU backend building:
.. code-block:: console
$ pip install --upgrade pip
$ pip install cmake>=3.26 wheel packaging ninja "setuptools-scm>=8" numpy
$ pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
- Finally, build and install vLLM CPU backend:
.. code-block:: console
$ VLLM_TARGET_DEVICE=cpu python setup.py install
.. note::
- AVX512_BF16 is an extension ISA provides native BF16 data type conversion and vector product instructions, will brings some performance improvement compared with pure AVX512. The CPU backend build script will check the host CPU flags to determine whether to enable AVX512_BF16.
- If you want to force enable AVX512_BF16 for the cross-compilation, please set environment variable VLLM_CPU_AVX512BF16=1 before the building.
.. _env_intro:
Related runtime environment variables
-------------------------------------
- ``VLLM_CPU_KVCACHE_SPACE``: specify the KV Cache size (e.g, ``VLLM_CPU_KVCACHE_SPACE=40`` means 40 GB space for KV cache), larger setting will allow vLLM running more requests in parallel. This parameter should be set based on the hardware configuration and memory management pattern of users.
- ``VLLM_CPU_OMP_THREADS_BIND``: specify the CPU cores dedicated to the OpenMP threads. For example, ``VLLM_CPU_OMP_THREADS_BIND=0-31`` means there will be 32 OpenMP threads bound on 0-31 CPU cores. ``VLLM_CPU_OMP_THREADS_BIND=0-31|32-63`` means there will be 2 tensor parallel processes, 32 OpenMP threads of rank0 are bound on 0-31 CPU cores, and the OpenMP threads of rank1 are bound on 32-63 CPU cores.
.. _ipex_guidance:
Intel Extension for PyTorch
---------------------------
- `Intel Extension for PyTorch (IPEX) <https://github.com/intel/intel-extension-for-pytorch>`_ extends PyTorch with up-to-date features optimizations for an extra performance boost on Intel hardware.
.. _cpu_backend_performance_tips:
Performance tips
-----------------
- We highly recommend to use TCMalloc for high performance memory allocation and better cache locality. For example, on Ubuntu 22.4, you can run:
.. code-block:: console
$ sudo apt-get install libtcmalloc-minimal4 # install TCMalloc library
$ find / -name *libtcmalloc* # find the dynamic link library path
$ export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libtcmalloc_minimal.so.4:$LD_PRELOAD # prepend the library to LD_PRELOAD
$ python examples/offline_inference.py # run vLLM
- When using the online serving, it is recommended to reserve 1-2 CPU cores for the serving framework to avoid CPU oversubscription. For example, on a platform with 32 physical CPU cores, reserving CPU 30 and 31 for the framework and using CPU 0-29 for OpenMP:
.. code-block:: console
$ export VLLM_CPU_KVCACHE_SPACE=40
$ export VLLM_CPU_OMP_THREADS_BIND=0-29
$ vllm serve facebook/opt-125m
- If using vLLM CPU backend on a machine with hyper-threading, it is recommended to bind only one OpenMP thread on each physical CPU core using ``VLLM_CPU_OMP_THREADS_BIND``. On a hyper-threading enabled platform with 16 logical CPU cores / 8 physical CPU cores:
.. code-block:: console
$ lscpu -e # check the mapping between logical CPU cores and physical CPU cores
# The "CPU" column means the logical CPU core IDs, and the "CORE" column means the physical core IDs. On this platform, two logical cores are sharing one physical core.
CPU NODE SOCKET CORE L1d:L1i:L2:L3 ONLINE MAXMHZ MINMHZ MHZ
0 0 0 0 0:0:0:0 yes 2401.0000 800.0000 800.000
1 0 0 1 1:1:1:0 yes 2401.0000 800.0000 800.000
2 0 0 2 2:2:2:0 yes 2401.0000 800.0000 800.000
3 0 0 3 3:3:3:0 yes 2401.0000 800.0000 800.000
4 0 0 4 4:4:4:0 yes 2401.0000 800.0000 800.000
5 0 0 5 5:5:5:0 yes 2401.0000 800.0000 800.000
6 0 0 6 6:6:6:0 yes 2401.0000 800.0000 800.000
7 0 0 7 7:7:7:0 yes 2401.0000 800.0000 800.000
8 0 0 0 0:0:0:0 yes 2401.0000 800.0000 800.000
9 0 0 1 1:1:1:0 yes 2401.0000 800.0000 800.000
10 0 0 2 2:2:2:0 yes 2401.0000 800.0000 800.000
11 0 0 3 3:3:3:0 yes 2401.0000 800.0000 800.000
12 0 0 4 4:4:4:0 yes 2401.0000 800.0000 800.000
13 0 0 5 5:5:5:0 yes 2401.0000 800.0000 800.000
14 0 0 6 6:6:6:0 yes 2401.0000 800.0000 800.000
15 0 0 7 7:7:7:0 yes 2401.0000 800.0000 800.000
# On this platform, it is recommend to only bind openMP threads on logical CPU cores 0-7 or 8-15
$ export VLLM_CPU_OMP_THREADS_BIND=0-7
$ python examples/offline_inference.py
- If using vLLM CPU backend on a multi-socket machine with NUMA, be aware to set CPU cores using ``VLLM_CPU_OMP_THREADS_BIND`` to avoid cross NUMA node memory access.
CPU Backend Considerations
--------------------------
- The CPU backend significantly differs from the GPU backend since the vLLM architecture was originally optimized for GPU use. A number of optimizations are needed to enhance its performance.
- Decouple the HTTP serving components from the inference components. In a GPU backend configuration, the HTTP serving and tokenization tasks operate on the CPU, while inference runs on the GPU, which typically does not pose a problem. However, in a CPU-based setup, the HTTP serving and tokenization can cause significant context switching and reduced cache efficiency. Therefore, it is strongly recommended to segregate these two components for improved performance.
- On CPU based setup with NUMA enabled, the memory access performance may be largely impacted by the `topology <https://github.com/intel/intel-extension-for-pytorch/blob/main/docs/tutorials/performance_tuning/tuning_guide.md#non-uniform-memory-access-numa>`_. For NUMA architecture, two optimizations are to recommended: Tensor Parallel or Data Parallel.
* Using Tensor Parallel for a latency constraints deployment: following GPU backend design, a Megatron-LM's parallel algorithm will be used to shard the model, based on the number of NUMA nodes (e.g. TP = 2 for a two NUMA node system). With `TP feature on CPU <https://github.com/vllm-project/vllm/pull/6125>`_ merged, Tensor Parallel is supported for serving and offline inferencing. In general each NUMA node is treated as one GPU card. Below is the example script to enable Tensor Parallel = 2 for serving:
.. code-block:: console
$ VLLM_CPU_KVCACHE_SPACE=40 VLLM_CPU_OMP_THREADS_BIND="0-31|32-63" vllm serve meta-llama/Llama-2-7b-chat-hf -tp=2 --distributed-executor-backend mp
* Using Data Parallel for maximum throughput: to launch an LLM serving endpoint on each NUMA node along with one additional load balancer to dispatch the requests to those endpoints. Common solutions like `Nginx <../serving/deploying_with_nginx.html>`_ or HAProxy are recommended. Anyscale Ray project provides the feature on LLM `serving <https://docs.ray.io/en/latest/serve/index.html>`_. Here is the example to setup a scalable LLM serving with `Ray Serve <https://github.com/intel/llm-on-ray/blob/main/docs/setup.md>`_.

199
docs/source/getting_started/debugging.md Normal file
View File

@ -0,0 +1,199 @@
(debugging)=
# Debugging Tips
This document outlines some debugging strategies you can consider. If you think you've discovered a bug, please [search existing issues](https://github.com/vllm-project/vllm/issues?q=is%3Aissue) first to see if it has already been reported. If not, please [file a new issue](https://github.com/vllm-project/vllm/issues/new/choose), providing as much relevant information as possible.
```{note}
Once you've debugged a problem, remember to turn off any debugging environment variables defined, or simply start a new shell to avoid being affected by lingering debugging settings. Otherwise, the system might be slow with debugging functionalities left activated.
```
## Hangs downloading a model
If the model isn't already downloaded to disk, vLLM will download it from the internet which can take time and depend on your internet connection.
It's recommended to download the model first using the [huggingface-cli](https://huggingface.co/docs/huggingface_hub/en/guides/cli) and passing the local path to the model to vLLM. This way, you can isolate the issue.
## Hangs loading a model from disk
If the model is large, it can take a long time to load it from disk. Pay attention to where you store the model. Some clusters have shared filesystems across nodes, e.g. a distributed filesystem or a network filesystem, which can be slow.
It'd be better to store the model in a local disk. Additionally, have a look at the CPU memory usage, when the model is too large it might take a lot of CPU memory, slowing down the operating system because it needs to frequently swap between disk and memory.
```{note}
To isolate the model downloading and loading issue, you can use the `--load-format dummy` argument to skip loading the model weights. This way, you can check if the model downloading and loading is the bottleneck.
```
## Model is too large
If the model is too large to fit in a single GPU, you might want to [consider tensor parallelism](https://docs.vllm.ai/en/latest/serving/distributed_serving.html#distributed-inference-and-serving) to split the model across multiple GPUs. In that case, every process will read the whole model and split it into chunks, which makes the disk reading time even longer (proportional to the size of tensor parallelism). You can convert the model checkpoint to a sharded checkpoint using [this example](https://docs.vllm.ai/en/latest/getting_started/examples/save_sharded_state.html) . The conversion process might take some time, but later you can load the sharded checkpoint much faster. The model loading time should remain constant regardless of the size of tensor parallelism.
## Enable more logging
If other strategies don't solve the problem, it's likely that the vLLM instance is stuck somewhere. You can use the following environment variables to help debug the issue:
- `export VLLM_LOGGING_LEVEL=DEBUG` to turn on more logging.
- `export CUDA_LAUNCH_BLOCKING=1` to identify which CUDA kernel is causing the problem.
- `export NCCL_DEBUG=TRACE` to turn on more logging for NCCL.
- `export VLLM_TRACE_FUNCTION=1` to record all function calls for inspection in the log files to tell which function crashes or hangs.
## Incorrect network setup
The vLLM instance cannot get the correct IP address if you have a complicated network config. You can find a log such as `DEBUG 06-10 21:32:17 parallel_state.py:88] world_size=8 rank=0 local_rank=0 distributed_init_method=tcp://xxx.xxx.xxx.xxx:54641 backend=nccl` and the IP address should be the correct one.
If it's not, override the IP address using the environment variable `export VLLM_HOST_IP=<your_ip_address>`.
You might also need to set `export NCCL_SOCKET_IFNAME=<your_network_interface>` and `export GLOO_SOCKET_IFNAME=<your_network_interface>` to specify the network interface for the IP address.
## Error near `self.graph.replay()`
If vLLM crashes and the error trace captures it somewhere around `self.graph.replay()` in `vllm/worker/model_runner.py`, it is a CUDA error inside CUDAGraph.
To identify the particular CUDA operation that causes the error, you can add `--enforce-eager` to the command line, or `enforce_eager=True` to the {class}`~vllm.LLM` class to disable the CUDAGraph optimization and isolate the exact CUDA operation that causes the error.
## Incorrect hardware/driver
If GPU/CPU communication cannot be established, you can use the following Python script and follow the instructions below to confirm whether the GPU/CPU communication is working correctly.
```python
# Test PyTorch NCCL
import torch
import torch.distributed as dist
dist.init_process_group(backend="nccl")
local_rank = dist.get_rank() % torch.cuda.device_count()
torch.cuda.set_device(local_rank)
data = torch.FloatTensor([1,] * 128).to("cuda")
dist.all_reduce(data, op=dist.ReduceOp.SUM)
torch.cuda.synchronize()
value = data.mean().item()
world_size = dist.get_world_size()
assert value == world_size, f"Expected {world_size}, got {value}"
print("PyTorch NCCL is successful!")
# Test PyTorch GLOO
gloo_group = dist.new_group(ranks=list(range(world_size)), backend="gloo")
cpu_data = torch.FloatTensor([1,] * 128)
dist.all_reduce(cpu_data, op=dist.ReduceOp.SUM, group=gloo_group)
value = cpu_data.mean().item()
assert value == world_size, f"Expected {world_size}, got {value}"
print("PyTorch GLOO is successful!")
if world_size <= 1:
exit()
# Test vLLM NCCL, with cuda graph
from vllm.distributed.device_communicators.pynccl import PyNcclCommunicator
pynccl = PyNcclCommunicator(group=gloo_group, device=local_rank)
# pynccl is enabled by default for 0.6.5+,
# but for 0.6.4 and below, we need to enable it manually.
# keep the code for backward compatibility when because people
# prefer to read the latest documentation.
pynccl.disabled = False
s = torch.cuda.Stream()
with torch.cuda.stream(s):
data.fill_(1)
pynccl.all_reduce(data, stream=s)
value = data.mean().item()
assert value == world_size, f"Expected {world_size}, got {value}"
print("vLLM NCCL is successful!")
g = torch.cuda.CUDAGraph()
with torch.cuda.graph(cuda_graph=g, stream=s):
pynccl.all_reduce(data, stream=torch.cuda.current_stream())
data.fill_(1)
g.replay()
torch.cuda.current_stream().synchronize()
value = data.mean().item()
assert value == world_size, f"Expected {world_size}, got {value}"
print("vLLM NCCL with cuda graph is successful!")
dist.destroy_process_group(gloo_group)
dist.destroy_process_group()
```
If you are testing with a single node, adjust `--nproc-per-node` to the number of GPUs you want to use:
```console
$ NCCL_DEBUG=TRACE torchrun --nproc-per-node=<number-of-GPUs> test.py
```
If you are testing with multi-nodes, adjust `--nproc-per-node` and `--nnodes` according to your setup and set `MASTER_ADDR` to the correct IP address of the master node, reachable from all nodes. Then, run:
```console
$ NCCL_DEBUG=TRACE torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=$MASTER_ADDR test.py
```
If the script runs successfully, you should see the message `sanity check is successful!`.
If the test script hangs or crashes, usually it means the hardware/drivers are broken in some sense. You should try to contact your system administrator or hardware vendor for further assistance. As a common workaround, you can try to tune some NCCL environment variables, such as `export NCCL_P2P_DISABLE=1` to see if it helps. Please check [their documentation](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html) for more information. Please only use these environment variables as a temporary workaround, as they might affect the performance of the system. The best solution is still to fix the hardware/drivers so that the test script can run successfully.
```{note}
A multi-node environment is more complicated than a single-node one. If you see errors such as `torch.distributed.DistNetworkError`, it is likely that the network/DNS setup is incorrect. In that case, you can manually assign node rank and specify the IP via command line arguments:
- In the first node, run `NCCL_DEBUG=TRACE torchrun --nnodes 2 --nproc-per-node=2 --node-rank 0 --master_addr $MASTER_ADDR test.py`.
- In the second node, run `NCCL_DEBUG=TRACE torchrun --nnodes 2 --nproc-per-node=2 --node-rank 1 --master_addr $MASTER_ADDR test.py`.
Adjust `--nproc-per-node`, `--nnodes`, and `--node-rank` according to your setup, being sure to execute different commands (with different `--node-rank`) on different nodes.
```
## Python multiprocessing
### `RuntimeError` Exception
If you have seen a warning in your logs like this:
```console
WARNING 12-11 14:50:37 multiproc_worker_utils.py:281] CUDA was previously
initialized. We must use the `spawn` multiprocessing start method. Setting
VLLM_WORKER_MULTIPROC_METHOD to 'spawn'. See
https://docs.vllm.ai/en/latest/getting_started/debugging.html#python-multiprocessing
for more information.
```
or an error from Python that looks like this:
```console
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.
This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:
if __name__ == '__main__':
freeze_support()
...
The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.
To fix this issue, refer to the "Safe importing of main module"
section in https://docs.python.org/3/library/multiprocessing.html
```
then you must update your Python code to guard usage of `vllm` behind a `if
__name__ == '__main__':` block. For example, instead of this:
```python
import vllm
llm = vllm.LLM(...)
```
try this instead:
```python
if __name__ == '__main__':
import vllm
llm = vllm.LLM(...)
```
## Known Issues
- In `v0.5.2`, `v0.5.3`, and `v0.5.3.post1`, there is a bug caused by [zmq](https://github.com/zeromq/pyzmq/issues/2000) , which can occasionally cause vLLM to hang depending on the machine configuration. The solution is to upgrade to the latest version of `vllm` to include the [fix](https://github.com/vllm-project/vllm/pull/6759).
- To circumvent a NCCL [bug](https://github.com/NVIDIA/nccl/issues/1234) , all vLLM processes will set an environment variable ``NCCL_CUMEM_ENABLE=0`` to disable NCCL's ``cuMem`` allocator. It does not affect performance but only gives memory benefits. When external processes want to set up a NCCL connection with vLLM's processes, they should also set this environment variable, otherwise, inconsistent environment setup will cause NCCL to hang or crash, as observed in the [RLHF integration](https://github.com/OpenRLHF/OpenRLHF/pull/604) and the [discussion](https://github.com/vllm-project/vllm/issues/5723#issuecomment-2554389656) .

203
docs/source/getting_started/debugging.rst
View File

@ -1,203 +0,0 @@
.. _debugging:
===============
Debugging Tips
===============
This document outlines some debugging strategies you can consider. If you think you've discovered a bug, please `search existing issues <https://github.com/vllm-project/vllm/issues?q=is%3Aissue>`_ first to see if it has already been reported. If not, please `file a new issue <https://github.com/vllm-project/vllm/issues/new/choose>`_, providing as much relevant information as possible.
.. note::
Once you've debugged a problem, remember to turn off any debugging environment variables defined, or simply start a new shell to avoid being affected by lingering debugging settings. Otherwise, the system might be slow with debugging functionalities left activated.
Hangs downloading a model
----------------------------------------
If the model isn't already downloaded to disk, vLLM will download it from the internet which can take time and depend on your internet connection.
It's recommended to download the model first using the `huggingface-cli <https://huggingface.co/docs/huggingface_hub/en/guides/cli>`_ and passing the local path to the model to vLLM. This way, you can isolate the issue.
Hangs loading a model from disk
----------------------------------------
If the model is large, it can take a long time to load it from disk. Pay attention to where you store the model. Some clusters have shared filesystems across nodes, e.g. a distributed filesystem or a network filesystem, which can be slow.
It'd be better to store the model in a local disk. Additionally, have a look at the CPU memory usage, when the model is too large it might take a lot of CPU memory, slowing down the operating system because it needs to frequently swap between disk and memory.
.. note::
To isolate the model downloading and loading issue, you can use the ``--load-format dummy`` argument to skip loading the model weights. This way, you can check if the model downloading and loading is the bottleneck.
Model is too large
----------------------------------------
If the model is too large to fit in a single GPU, you might want to `consider tensor parallelism <https://docs.vllm.ai/en/latest/serving/distributed_serving.html#distributed-inference-and-serving>`_ to split the model across multiple GPUs. In that case, every process will read the whole model and split it into chunks, which makes the disk reading time even longer (proportional to the size of tensor parallelism). You can convert the model checkpoint to a sharded checkpoint using `this example <https://docs.vllm.ai/en/latest/getting_started/examples/save_sharded_state.html>`_ . The conversion process might take some time, but later you can load the sharded checkpoint much faster. The model loading time should remain constant regardless of the size of tensor parallelism.
Enable more logging
----------------------------------------
If other strategies don't solve the problem, it's likely that the vLLM instance is stuck somewhere. You can use the following environment variables to help debug the issue:
- ``export VLLM_LOGGING_LEVEL=DEBUG`` to turn on more logging.
- ``export CUDA_LAUNCH_BLOCKING=1`` to identify which CUDA kernel is causing the problem.
- ``export NCCL_DEBUG=TRACE`` to turn on more logging for NCCL.
- ``export VLLM_TRACE_FUNCTION=1`` to record all function calls for inspection in the log files to tell which function crashes or hangs.
Incorrect network setup
----------------------------------------
The vLLM instance cannot get the correct IP address if you have a complicated network config. You can find a log such as ``DEBUG 06-10 21:32:17 parallel_state.py:88] world_size=8 rank=0 local_rank=0 distributed_init_method=tcp://xxx.xxx.xxx.xxx:54641 backend=nccl`` and the IP address should be the correct one.
If it's not, override the IP address using the environment variable ``export VLLM_HOST_IP=<your_ip_address>``.
You might also need to set ``export NCCL_SOCKET_IFNAME=<your_network_interface>`` and ``export GLOO_SOCKET_IFNAME=<your_network_interface>`` to specify the network interface for the IP address.
Error near ``self.graph.replay()``
----------------------------------------
If vLLM crashes and the error trace captures it somewhere around ``self.graph.replay()`` in ``vllm/worker/model_runner.py``, it is a CUDA error inside CUDAGraph.
To identify the particular CUDA operation that causes the error, you can add ``--enforce-eager`` to the command line, or ``enforce_eager=True`` to the :class:`~vllm.LLM` class to disable the CUDAGraph optimization and isolate the exact CUDA operation that causes the error.
Incorrect hardware/driver
----------------------------------------
If GPU/CPU communication cannot be established, you can use the following Python script and follow the instructions below to confirm whether the GPU/CPU communication is working correctly.
.. code-block:: python
# Test PyTorch NCCL
import torch
import torch.distributed as dist
dist.init_process_group(backend="nccl")
local_rank = dist.get_rank() % torch.cuda.device_count()
torch.cuda.set_device(local_rank)
data = torch.FloatTensor([1,] * 128).to("cuda")
dist.all_reduce(data, op=dist.ReduceOp.SUM)
torch.cuda.synchronize()
value = data.mean().item()
world_size = dist.get_world_size()
assert value == world_size, f"Expected {world_size}, got {value}"
print("PyTorch NCCL is successful!")
# Test PyTorch GLOO
gloo_group = dist.new_group(ranks=list(range(world_size)), backend="gloo")
cpu_data = torch.FloatTensor([1,] * 128)
dist.all_reduce(cpu_data, op=dist.ReduceOp.SUM, group=gloo_group)
value = cpu_data.mean().item()
assert value == world_size, f"Expected {world_size}, got {value}"
print("PyTorch GLOO is successful!")
if world_size <= 1:
exit()
# Test vLLM NCCL, with cuda graph
from vllm.distributed.device_communicators.pynccl import PyNcclCommunicator
pynccl = PyNcclCommunicator(group=gloo_group, device=local_rank)
# pynccl is enabled by default for 0.6.5+,
# but for 0.6.4 and below, we need to enable it manually.
# keep the code for backward compatibility when because people
# prefer to read the latest documentation.
pynccl.disabled = False
s = torch.cuda.Stream()
with torch.cuda.stream(s):
data.fill_(1)
pynccl.all_reduce(data, stream=s)
value = data.mean().item()
assert value == world_size, f"Expected {world_size}, got {value}"
print("vLLM NCCL is successful!")
g = torch.cuda.CUDAGraph()
with torch.cuda.graph(cuda_graph=g, stream=s):
pynccl.all_reduce(data, stream=torch.cuda.current_stream())
data.fill_(1)
g.replay()
torch.cuda.current_stream().synchronize()
value = data.mean().item()
assert value == world_size, f"Expected {world_size}, got {value}"
print("vLLM NCCL with cuda graph is successful!")
dist.destroy_process_group(gloo_group)
dist.destroy_process_group()
If you are testing with a single node, adjust ``--nproc-per-node`` to the number of GPUs you want to use:
.. code-block:: console
$ NCCL_DEBUG=TRACE torchrun --nproc-per-node=<number-of-GPUs> test.py
If you are testing with multi-nodes, adjust ``--nproc-per-node`` and ``--nnodes`` according to your setup and set ``MASTER_ADDR`` to the correct IP address of the master node, reachable from all nodes. Then, run:
.. code-block:: console
$ NCCL_DEBUG=TRACE torchrun --nnodes 2 --nproc-per-node=2 --rdzv_backend=c10d --rdzv_endpoint=$MASTER_ADDR test.py
If the script runs successfully, you should see the message ``sanity check is successful!``.
If the test script hangs or crashes, usually it means the hardware/drivers are broken in some sense. You should try to contact your system administrator or hardware vendor for further assistance. As a common workaround, you can try to tune some NCCL environment variables, such as ``export NCCL_P2P_DISABLE=1`` to see if it helps. Please check `their documentation <https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/env.html>`__ for more information. Please only use these environment variables as a temporary workaround, as they might affect the performance of the system. The best solution is still to fix the hardware/drivers so that the test script can run successfully.
.. note::
A multi-node environment is more complicated than a single-node one. If you see errors such as ``torch.distributed.DistNetworkError``, it is likely that the network/DNS setup is incorrect. In that case, you can manually assign node rank and specify the IP via command line arguments:
- In the first node, run ``NCCL_DEBUG=TRACE torchrun --nnodes 2 --nproc-per-node=2 --node-rank 0 --master_addr $MASTER_ADDR test.py``.
- In the second node, run ``NCCL_DEBUG=TRACE torchrun --nnodes 2 --nproc-per-node=2 --node-rank 1 --master_addr $MASTER_ADDR test.py``.
Adjust ``--nproc-per-node``, ``--nnodes``, and ``--node-rank`` according to your setup, being sure to execute different commands (with different ``--node-rank``) on different nodes.
Python multiprocessing
----------------------
`RuntimeError` Exception
^^^^^^^^^^^^^^^^^^^^^^^^
If you have seen a warning in your logs like this:
.. code-block:: console
WARNING 12-11 14:50:37 multiproc_worker_utils.py:281] CUDA was previously
initialized. We must use the `spawn` multiprocessing start method. Setting
VLLM_WORKER_MULTIPROC_METHOD to 'spawn'. See
https://docs.vllm.ai/en/latest/getting_started/debugging.html#python-multiprocessing
for more information.
or an error from Python that looks like this:
.. code-block:: console
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.
This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:
if __name__ == '__main__':
freeze_support()
...
The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.
To fix this issue, refer to the "Safe importing of main module"
section in https://docs.python.org/3/library/multiprocessing.html
then you must update your Python code to guard usage of ``vllm`` behind a ``if
__name__ == '__main__':`` block. For example, instead of this:
.. code-block:: python
import vllm
llm = vllm.LLM(...)
try this instead:
.. code-block:: python
if __name__ == '__main__':
import vllm
llm = vllm.LLM(...)
Known Issues
----------------------------------------
- In ``v0.5.2``, ``v0.5.3``, and ``v0.5.3.post1``, there is a bug caused by `zmq <https://github.com/zeromq/pyzmq/issues/2000>`_ , which can occasionally cause vLLM to hang depending on the machine configuration. The solution is to upgrade to the latest version of ``vllm`` to include the `fix <https://github.com/vllm-project/vllm/pull/6759>`_.
- To circumvent a NCCL `bug <https://github.com/NVIDIA/nccl/issues/1234>`__ , all vLLM processes will set an environment variable ``NCCL_CUMEM_ENABLE=0`` to disable NCCL's ``cuMem`` allocator. It does not affect performance but only gives memory benefits. When external processes want to set up a NCCL connection with vLLM's processes, they should also set this environment variable, otherwise, inconsistent environment setup will cause NCCL to hang or crash, as observed in `the RLHF integration <https://github.com/OpenRLHF/OpenRLHF/pull/604>`__ and the `discussion <https://github.com/vllm-project/vllm/issues/5723#issuecomment-2554389656>`__ .

8
docs/source/getting_started/examples/examples_index.template.md Normal file
View File

@ -0,0 +1,8 @@
# Examples
```{toctree}
:maxdepth: 1
:caption: Scripts
%EXAMPLE_DOCS%
```

8
docs/source/getting_started/examples/examples_index.template.rst
View File

@ -1,8 +0,0 @@
Examples
=================================
.. toctree::
:maxdepth: 1
:caption: Scripts
%EXAMPLE_DOCS%

388
docs/source/getting_started/gaudi-installation.md Normal file
View File

@ -0,0 +1,388 @@
# Installation with Intel® Gaudi® AI Accelerators
This README provides instructions on running vLLM with Intel Gaudi devices.
## Requirements and Installation
Please follow the instructions provided in the [Gaudi Installation
Guide](https://docs.habana.ai/en/latest/Installation_Guide/index.html)
to set up the execution environment. To achieve the best performance,
please follow the methods outlined in the [Optimizing Training Platform
Guide](https://docs.habana.ai/en/latest/PyTorch/Model_Optimization_PyTorch/Optimization_in_Training_Platform.html).
### Requirements
- OS: Ubuntu 22.04 LTS
- Python: 3.10
- Intel Gaudi accelerator
- Intel Gaudi software version 1.18.0
### Quick start using Dockerfile
```console
$ docker build -f Dockerfile.hpu -t vllm-hpu-env .
$ docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --rm vllm-hpu-env
```
```{tip}
If you're observing the following error: `docker: Error response from daemon: Unknown runtime specified habana.`, please refer to "Install Using Containers" section of [Intel Gaudi Software Stack and Driver Installation](https://docs.habana.ai/en/v1.18.0/Installation_Guide/Bare_Metal_Fresh_OS.html). Make sure you have `habana-container-runtime` package installed and that `habana` container runtime is registered.
```
### Build from source
#### Environment verification
To verify that the Intel Gaudi software was correctly installed, run:
```console
$ hl-smi # verify that hl-smi is in your PATH and each Gaudi accelerator is visible
$ apt list --installed | grep habana # verify that habanalabs-firmware-tools, habanalabs-graph, habanalabs-rdma-core, habanalabs-thunk and habanalabs-container-runtime are installed
$ pip list | grep habana # verify that habana-torch-plugin, habana-torch-dataloader, habana-pyhlml and habana-media-loader are installed
$ pip list | grep neural # verify that neural_compressor is installed
```
Refer to [Intel Gaudi Software Stack
Verification](https://docs.habana.ai/en/latest/Installation_Guide/SW_Verification.html#platform-upgrade)
for more details.
#### Run Docker Image
It is highly recommended to use the latest Docker image from Intel Gaudi
vault. Refer to the [Intel Gaudi
documentation](https://docs.habana.ai/en/latest/Installation_Guide/Bare_Metal_Fresh_OS.html#pull-prebuilt-containers)
for more details.
Use the following commands to run a Docker image:
```console
$ docker pull vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.0:latest
$ docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.0:latest
```
#### Build and Install vLLM
To build and install vLLM from source, run:
```console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ python setup.py develop
```
Currently, the latest features and performance optimizations are developed in Gaudi's [vLLM-fork](https://github.com/HabanaAI/vllm-fork) and we periodically upstream them to vLLM main repo. To install latest [HabanaAI/vLLM-fork](https://github.com/HabanaAI/vllm-fork), run the following:
```console
$ git clone https://github.com/HabanaAI/vllm-fork.git
$ cd vllm-fork
$ git checkout habana_main
$ python setup.py develop
```
## Supported Features
- [Offline batched
inference](https://docs.vllm.ai/en/latest/getting_started/quickstart.html#offline-batched-inference)
- Online inference via [OpenAI-Compatible
Server](https://docs.vllm.ai/en/latest/getting_started/quickstart.html#openai-compatible-server)
- HPU autodetection - no need to manually select device within vLLM
- Paged KV cache with algorithms enabled for Intel Gaudi accelerators
- Custom Intel Gaudi implementations of Paged Attention, KV cache ops,
prefill attention, Root Mean Square Layer Normalization, Rotary
Positional Encoding
- Tensor parallelism support for multi-card inference
- Inference with [HPU Graphs](https://docs.habana.ai/en/latest/PyTorch/Inference_on_PyTorch/Inference_Using_HPU_Graphs.html)
for accelerating low-batch latency and throughput
- Attention with Linear Biases (ALiBi)
## Unsupported Features
- Beam search
- LoRA adapters
- Quantization
- Prefill chunking (mixed-batch inferencing)
## Supported Configurations
The following configurations have been validated to be function with
Gaudi2 devices. Configurations that are not listed may or may not work.
- [meta-llama/Llama-2-7b](https://huggingface.co/meta-llama/Llama-2-7b)
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf)
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B)
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct)
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3.1-8B](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B)
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct)
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- [meta-llama/Llama-2-70b](https://huggingface.co/meta-llama/Llama-2-70b)
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- [meta-llama/Llama-2-70b-chat-hf](https://huggingface.co/meta-llama/Llama-2-70b-chat-hf)
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3-70B](https://huggingface.co/meta-llama/Meta-Llama-3-70B)
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct)
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3.1-70B](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B)
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- [meta-llama/Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct)
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
## Performance Tuning
### Execution modes
Currently in vLLM for HPU we support four execution modes, depending on selected HPU PyTorch Bridge backend (via `PT_HPU_LAZY_MODE` environment variable), and `--enforce-eager` flag.
```{eval-rst}
.. list-table:: vLLM execution modes
:widths: 25 25 50
:header-rows: 1
* - ``PT_HPU_LAZY_MODE``
- ``enforce_eager``
- execution mode
* - 0
- 0
- torch.compile
* - 0
- 1
- PyTorch eager mode
* - 1
- 0
- HPU Graphs
* - 1
- 1
- PyTorch lazy mode
```
```{warning}
In 1.18.0, all modes utilizing `PT_HPU_LAZY_MODE=0` are highly experimental and should be only used for validating functional correctness. Their performance will be improved in the next releases. For obtaining the best performance in 1.18.0, please use HPU Graphs, or PyTorch lazy mode.
```
### Bucketing mechanism
Intel Gaudi accelerators work best when operating on models with fixed tensor shapes. [Intel Gaudi Graph Compiler](https://docs.habana.ai/en/latest/Gaudi_Overview/Intel_Gaudi_Software_Suite.html#graph-compiler-and-runtime) is responsible for generating optimized binary code that implements the given model topology on Gaudi. In its default configuration, the produced binary code may be heavily dependent on input and output tensor shapes, and can require graph recompilation when encountering differently shaped tensors within the same topology. While the resulting binaries utilize Gaudi efficiently, the compilation itself may introduce a noticeable overhead in end-to-end execution.
In a dynamic inference serving scenario, there is a need to minimize the number of graph compilations and reduce the risk of graph compilation occurring during server runtime. Currently it is achieved by "bucketing" model's forward pass across two dimensions - `batch_size` and `sequence_length`.
```{note}
Bucketing allows us to reduce the number of required graphs significantly, but it does not handle any graph compilation and device code generation - this is done in warmup and HPUGraph capture phase.
```
Bucketing ranges are determined with 3 parameters - `min`, `step` and `max`. They can be set separately for prompt and decode phase, and for batch size and sequence length dimension. These parameters can be observed in logs during vLLM startup:
```
INFO 08-01 21:37:59 hpu_model_runner.py:493] Prompt bucket config (min, step, max_warmup) bs:[1, 32, 4], seq:[128, 128, 1024]
INFO 08-01 21:37:59 hpu_model_runner.py:499] Generated 24 prompt buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024)]
INFO 08-01 21:37:59 hpu_model_runner.py:504] Decode bucket config (min, step, max_warmup) bs:[1, 128, 4], seq:[128, 128, 2048]
INFO 08-01 21:37:59 hpu_model_runner.py:509] Generated 48 decode buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (2, 1152), (2, 1280), (2, 1408), (2, 1536), (2, 1664), (2, 1792), (2, 1920), (2, 2048), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024), (4, 1152), (4, 1280), (4, 1408), (4, 1536), (4, 1664), (4, 1792), (4, 1920), (4, 2048)]
```
`min` determines the lowest value of the bucket. `step` determines the interval between buckets, and `max` determines the upper bound of the bucket. Furthermore, interval between `min` and `step` has special handling - `min` gets multiplied by consecutive powers of two, until `step` gets reached. We call this the ramp-up phase and it is used for handling lower batch sizes with minimum wastage, while allowing larger padding on larger batch sizes.
Example (with ramp-up)
```
min = 2, step = 32, max = 64
=> ramp_up = (2, 4, 8, 16)
=> stable = (32, 64)
=> buckets = ramp_up + stable => (2, 4, 8, 16, 32, 64)
```
Example (without ramp-up)
```
min = 128, step = 128, max = 512
=> ramp_up = ()
=> stable = (128, 256, 384, 512)
=> buckets = ramp_up + stable => (128, 256, 384, 512)
```
In the logged scenario, 24 buckets were generated for prompt (prefill) runs, and 48 buckets for decode runs. Each bucket corresponds to a separate optimized device binary for a given model with specified tensor shapes. Whenever a batch of requests is processed, it is padded across batch and sequence length dimension to the smallest possible bucket.
```{warning}
If a request exceeds maximum bucket size in any dimension, it will be processed without padding, and its processing may require a graph compilation, potentially significantly increasing end-to-end latency. The boundaries of the buckets are user-configurable via environment variables, and upper bucket boundaries can be increased to avoid such scenario.
```
As an example, if a request of 3 sequences, with max sequence length of 412 comes in to an idle vLLM server, it will be padded executed as `(4, 512)` prefill bucket, as `batch_size` (number of sequences) will be padded to 4 (closest batch_size dimension higher than 3), and max sequence length will be padded to 512 (closest sequence length dimension higher than 412). After prefill stage, it will be executed as `(4, 512)` decode bucket and will continue as that bucket until either batch dimension changes (due to request being finished) - in which case it will become a `(2, 512)` bucket, or context length increases above 512 tokens, in which case it will become `(4, 640)` bucket.
```{note}
Bucketing is transparent to a client - padding in sequence length dimension is never returned to the client, and padding in batch dimension does not create new requests.
```
### Warmup
Warmup is an optional, but highly recommended step occurring before vLLM server starts listening. It executes a forward pass for each bucket with dummy data. The goal is to pre-compile all graphs and not incur any graph compilation overheads within bucket boundaries during server runtime. Each warmup step is logged during vLLM startup:
```
INFO 08-01 22:26:47 hpu_model_runner.py:1066] [Warmup][Prompt][1/24] batch_size:4 seq_len:1024 free_mem:79.16 GiB
INFO 08-01 22:26:47 hpu_model_runner.py:1066] [Warmup][Prompt][2/24] batch_size:4 seq_len:896 free_mem:55.43 GiB
INFO 08-01 22:26:48 hpu_model_runner.py:1066] [Warmup][Prompt][3/24] batch_size:4 seq_len:768 free_mem:55.43 GiB
...
INFO 08-01 22:26:59 hpu_model_runner.py:1066] [Warmup][Prompt][24/24] batch_size:1 seq_len:128 free_mem:55.43 GiB
INFO 08-01 22:27:00 hpu_model_runner.py:1066] [Warmup][Decode][1/48] batch_size:4 seq_len:2048 free_mem:55.43 GiB
INFO 08-01 22:27:00 hpu_model_runner.py:1066] [Warmup][Decode][2/48] batch_size:4 seq_len:1920 free_mem:55.43 GiB
INFO 08-01 22:27:01 hpu_model_runner.py:1066] [Warmup][Decode][3/48] batch_size:4 seq_len:1792 free_mem:55.43 GiB
...
INFO 08-01 22:27:16 hpu_model_runner.py:1066] [Warmup][Decode][47/48] batch_size:2 seq_len:128 free_mem:55.43 GiB
INFO 08-01 22:27:16 hpu_model_runner.py:1066] [Warmup][Decode][48/48] batch_size:1 seq_len:128 free_mem:55.43 GiB
```
This example uses the same buckets as in *Bucketing mechanism* section. Each output line corresponds to execution of a single bucket. When bucket is executed for the first time, its graph is compiled and can be reused later on, skipping further graph compilations.
```{tip}
Compiling all the buckets might take some time and can be turned off with `VLLM_SKIP_WARMUP=true` environment variable. Keep in mind that if you do that, you may face graph compilations once executing a given bucket for the first time. It is fine to disable warmup for development, but it's highly recommended to enable it in deployment.
```
### HPU Graph capture
[HPU Graphs](https://docs.habana.ai/en/latest/PyTorch/Inference_on_PyTorch/Inference_Using_HPU_Graphs.html) are currently the most performant execution method of vLLM on Intel Gaudi. When HPU Graphs are enabled, execution graphs will be traced (recorded) ahead of time (after performing warmup), to be later replayed during inference, significantly reducing host overheads. Recording can take large amounts of memory, which needs to be taken into account when allocating KV cache. Enabling HPU Graphs will impact the number of available KV cache blocks, but vLLM provides user-configurable variables to control memory management.
When HPU Graphs are being used, they share the common memory pool ("usable memory") as KV cache, determined by `gpu_memory_utilization` flag (`0.9` by default).
Before KV cache gets allocated, model weights are loaded onto the device, and a forward pass of the model is executed on dummy data, to estimate memory usage.
Only after that, `gpu_memory_utilization` flag is utilized - at its default value, will mark 90% of free device memory at that point as usable.
Next, KV cache gets allocated, model is warmed up, and HPU Graphs are captured.
Environment variable `VLLM_GRAPH_RESERVED_MEM` defines the ratio of memory reserved for HPU Graphs capture.
With its default value (`VLLM_GRAPH_RESERVED_MEM=0.1`), 10% of usable memory will be reserved for graph capture (later referred to as "usable graph memory"), and the remaining 90% will be utilized for KV cache.
Environment variable `VLLM_GRAPH_PROMPT_RATIO` determines the ratio of usable graph memory reserved for prefill and decode graphs. By default (`VLLM_GRAPH_PROMPT_RATIO=0.3`), both stages have equal memory constraints.
Lower value corresponds to less usable graph memory reserved for prefill stage, e.g. `VLLM_GRAPH_PROMPT_RATIO=0.2` will reserve 20% of usable graph memory for prefill graphs, and 80% of usable graph memory for decode graphs.
```{note}
`gpu_memory_utilization` does not correspond to the absolute memory usage across HPU. It specifies the memory margin after loading the model and performing a profile run. If device has 100 GiB of total memory, and 50 GiB of free memory after loading model weights and executing profiling run, `gpu_memory_utilization` at its default value will mark 90% of 50 GiB as usable, leaving 5 GiB of margin, regardless of total device memory.
```
User can also configure the strategy for capturing HPU Graphs for prompt and decode stages separately. Strategy affects the order of capturing graphs. There are two strategies implemented:
\- `max_bs` - graph capture queue will sorted in descending order by their batch sizes. Buckets with equal batch sizes are sorted by sequence length in ascending order (e.g. `(64, 128)`, `(64, 256)`, `(32, 128)`, `(32, 256)`, `(1, 128)`, `(1,256)`), default strategy for decode
\- `min_tokens` - graph capture queue will be sorted in ascending order by the number of tokens each graph processes (`batch_size*sequence_length`), default strategy for prompt
When there's large amount of requests pending, vLLM scheduler will attempt to fill the maximum batch size for decode as soon as possible. When a request is finished, decode batch size decreases. When that happens, vLLM will attempt to schedule a prefill iteration for requests in the waiting queue, to fill the decode batch size to its previous state. This means that in a full load scenario, decode batch size is often at its maximum, which makes large batch size HPU Graphs crucial to capture, as reflected by `max_bs` strategy. On the other hand, prefills will be executed most frequently with very low batch sizes (1-4), which is reflected in `min_tokens` strategy.
```{note}
`VLLM_GRAPH_PROMPT_RATIO` does not set a hard limit on memory taken by graphs for each stage (prefill and decode). vLLM will first attempt to use up entirety of usable prefill graph memory (usable graph memory * `VLLM_GRAPH_PROMPT_RATIO`) for capturing prefill HPU Graphs, next it will attempt do the same for decode graphs and usable decode graph memory pool. If one stage is fully captured, and there is unused memory left within usable graph memory pool, vLLM will attempt further graph capture for the other stage, until no more HPU Graphs can be captured without exceeding reserved memory pool. The behavior on that mechanism can be observed in the example below.
```
Each described step is logged by vLLM server, as follows (negative values correspond to memory being released):
```
INFO 08-02 17:37:44 hpu_model_runner.py:493] Prompt bucket config (min, step, max_warmup) bs:[1, 32, 4], seq:[128, 128, 1024]
INFO 08-02 17:37:44 hpu_model_runner.py:499] Generated 24 prompt buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024)]
INFO 08-02 17:37:44 hpu_model_runner.py:504] Decode bucket config (min, step, max_warmup) bs:[1, 128, 4], seq:[128, 128, 2048]
INFO 08-02 17:37:44 hpu_model_runner.py:509] Generated 48 decode buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (2, 1152), (2, 1280), (2, 1408), (2, 1536), (2, 1664), (2, 1792), (2, 1920), (2, 2048), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024), (4, 1152), (4, 1280), (4, 1408), (4, 1536), (4, 1664), (4, 1792), (4, 1920), (4, 2048)]
INFO 08-02 17:37:52 hpu_model_runner.py:430] Pre-loading model weights on hpu:0 took 14.97 GiB of device memory (14.97 GiB/94.62 GiB used) and 2.95 GiB of host memory (475.2 GiB/1007 GiB used)
INFO 08-02 17:37:52 hpu_model_runner.py:438] Wrapping in HPU Graph took 0 B of device memory (14.97 GiB/94.62 GiB used) and -252 KiB of host memory (475.2 GiB/1007 GiB used)
INFO 08-02 17:37:52 hpu_model_runner.py:442] Loading model weights took in total 14.97 GiB of device memory (14.97 GiB/94.62 GiB used) and 2.95 GiB of host memory (475.2 GiB/1007 GiB used)
INFO 08-02 17:37:54 hpu_worker.py:134] Model profiling run took 504 MiB of device memory (15.46 GiB/94.62 GiB used) and 180.9 MiB of host memory (475.4 GiB/1007 GiB used)
INFO 08-02 17:37:54 hpu_worker.py:158] Free device memory: 79.16 GiB, 39.58 GiB usable (gpu_memory_utilization=0.5), 15.83 GiB reserved for HPUGraphs (VLLM_GRAPH_RESERVED_MEM=0.4), 23.75 GiB reserved for KV cache
INFO 08-02 17:37:54 hpu_executor.py:85] # HPU blocks: 1519, # CPU blocks: 0
INFO 08-02 17:37:54 hpu_worker.py:190] Initializing cache engine took 23.73 GiB of device memory (39.2 GiB/94.62 GiB used) and -1.238 MiB of host memory (475.4 GiB/1007 GiB used)
INFO 08-02 17:37:54 hpu_model_runner.py:1066] [Warmup][Prompt][1/24] batch_size:4 seq_len:1024 free_mem:55.43 GiB
...
INFO 08-02 17:38:22 hpu_model_runner.py:1066] [Warmup][Decode][48/48] batch_size:1 seq_len:128 free_mem:55.43 GiB
INFO 08-02 17:38:22 hpu_model_runner.py:1159] Using 15.85 GiB/55.43 GiB of free device memory for HPUGraphs, 7.923 GiB for prompt and 7.923 GiB for decode (VLLM_GRAPH_PROMPT_RATIO=0.3)
INFO 08-02 17:38:22 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][1/24] batch_size:1 seq_len:128 free_mem:55.43 GiB
...
INFO 08-02 17:38:26 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][11/24] batch_size:1 seq_len:896 free_mem:48.77 GiB
INFO 08-02 17:38:27 hpu_model_runner.py:1066] [Warmup][Graph/Decode][1/48] batch_size:4 seq_len:128 free_mem:47.51 GiB
...
INFO 08-02 17:38:41 hpu_model_runner.py:1066] [Warmup][Graph/Decode][48/48] batch_size:1 seq_len:2048 free_mem:47.35 GiB
INFO 08-02 17:38:41 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][12/24] batch_size:4 seq_len:256 free_mem:47.35 GiB
INFO 08-02 17:38:42 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][13/24] batch_size:2 seq_len:512 free_mem:45.91 GiB
INFO 08-02 17:38:42 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][14/24] batch_size:1 seq_len:1024 free_mem:44.48 GiB
INFO 08-02 17:38:43 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][15/24] batch_size:2 seq_len:640 free_mem:43.03 GiB
INFO 08-02 17:38:43 hpu_model_runner.py:1128] Graph/Prompt captured:15 (62.5%) used_mem:14.03 GiB buckets:[(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (4, 128), (4, 256)]
INFO 08-02 17:38:43 hpu_model_runner.py:1128] Graph/Decode captured:48 (100.0%) used_mem:161.9 MiB buckets:[(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (2, 1152), (2, 1280), (2, 1408), (2, 1536), (2, 1664), (2, 1792), (2, 1920), (2, 2048), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024), (4, 1152), (4, 1280), (4, 1408), (4, 1536), (4, 1664), (4, 1792), (4, 1920), (4, 2048)]
INFO 08-02 17:38:43 hpu_model_runner.py:1206] Warmup finished in 49 secs, allocated 14.19 GiB of device memory
INFO 08-02 17:38:43 hpu_executor.py:91] init_cache_engine took 37.92 GiB of device memory (53.39 GiB/94.62 GiB used) and 57.86 MiB of host memory (475.4 GiB/1007 GiB used)
```
### Recommended vLLM Parameters
- We recommend running inference on Gaudi 2 with `block_size` of 128
for BF16 data type. Using default values (16, 32) might lead to
sub-optimal performance due to Matrix Multiplication Engine
under-utilization (see [Gaudi
Architecture](https://docs.habana.ai/en/latest/Gaudi_Overview/Gaudi_Architecture.html)).
- For max throughput on Llama 7B, we recommend running with batch size
of 128 or 256 and max context length of 2048 with HPU Graphs enabled.
If you encounter out-of-memory issues, see troubleshooting section.
### Environment variables
**Diagnostic and profiling knobs:**
- `VLLM_PROFILER_ENABLED`: if `true`, high level profiler will be enabled. Resulting JSON traces can be viewed in [perfetto.habana.ai](https://perfetto.habana.ai/#!/viewer). Disabled by default.
- `VLLM_HPU_LOG_STEP_GRAPH_COMPILATION`: if `true`, will log graph compilations per each vLLM engine step, only when there was any - highly recommended to use alongside `PT_HPU_METRICS_GC_DETAILS=1`. Disabled by default.
- `VLLM_HPU_LOG_STEP_GRAPH_COMPILATION_ALL`: if `true`, will log graph compilations per each vLLM engine step, always, even if there were none. Disabled by default.
- `VLLM_HPU_LOG_STEP_CPU_FALLBACKS`: if `true`, will log cpu fallbacks per each vLLM engine step, only when there was any. Disabled by default.
- `VLLM_HPU_LOG_STEP_CPU_FALLBACKS_ALL`: if `true`, will log cpu fallbacks per each vLLM engine step, always, even if there were none. Disabled by default.
**Performance tuning knobs:**
- `VLLM_SKIP_WARMUP`: if `true`, warmup will be skipped, `false` by default
- `VLLM_GRAPH_RESERVED_MEM`: percentage of memory dedicated for HPUGraph capture, `0.1` by default
- `VLLM_GRAPH_PROMPT_RATIO`: percentage of reserved graph memory dedicated for prompt graphs, `0.3` by default
- `VLLM_GRAPH_PROMPT_STRATEGY`: strategy determining order of prompt graph capture, `min_tokens` or `max_bs`, `min_tokens` by default
- `VLLM_GRAPH_DECODE_STRATEGY`: strategy determining order of decode graph capture, `min_tokens` or `max_bs`, `max_bs` by default
- `VLLM_{phase}_{dim}_BUCKET_{param}` - collection of 12 environment variables configuring ranges of bucketing mechanism
- `{phase}` is either `PROMPT` or `DECODE`
- `{dim}` is either `BS`, `SEQ` or `BLOCK`
- `{param}` is either `MIN`, `STEP` or `MAX`
- Default values:
- Prompt:
: - batch size min (`VLLM_PROMPT_BS_BUCKET_MIN`): `1`
- batch size step (`VLLM_PROMPT_BS_BUCKET_STEP`): `min(max_num_seqs, 32)`
- batch size max (`VLLM_PROMPT_BS_BUCKET_MAX`): `min(max_num_seqs, 64)`
- sequence length min (`VLLM_PROMPT_SEQ_BUCKET_MIN`): `block_size`
- sequence length step (`VLLM_PROMPT_SEQ_BUCKET_STEP`): `block_size`
- sequence length max (`VLLM_PROMPT_SEQ_BUCKET_MAX`): `max_model_len`
- Decode:
: - batch size min (`VLLM_DECODE_BS_BUCKET_MIN`): `1`
- batch size step (`VLLM_DECODE_BS_BUCKET_STEP`): `min(max_num_seqs, 32)`
- batch size max (`VLLM_DECODE_BS_BUCKET_MAX`): `max_num_seqs`
- sequence length min (`VLLM_DECODE_BLOCK_BUCKET_MIN`): `block_size`
- sequence length step (`VLLM_DECODE_BLOCK_BUCKET_STEP`): `block_size`
- sequence length max (`VLLM_DECODE_BLOCK_BUCKET_MAX`): `max(128, (max_num_seqs*max_model_len)/block_size)`
Additionally, there are HPU PyTorch Bridge environment variables impacting vLLM execution:
- `PT_HPU_LAZY_MODE`: if `0`, PyTorch Eager backend for Gaudi will be used, if `1` PyTorch Lazy backend for Gaudi will be used, `1` is default
- `PT_HPU_ENABLE_LAZY_COLLECTIVES`: required to be `true` for tensor parallel inference with HPU Graphs
## Troubleshooting: Tweaking HPU Graphs
If you experience device out-of-memory issues or want to attempt
inference at higher batch sizes, try tweaking HPU Graphs by following
the below:
- Tweak `gpu_memory_utilization` knob. It will decrease the
allocation of KV cache, leaving some headroom for capturing graphs
with larger batch size. By default `gpu_memory_utilization` is set
to 0.9. It attempts to allocate ~90% of HBM left for KV cache after
short profiling run. Note that decreasing reduces the number of KV
cache blocks you have available, and therefore reduces the effective
maximum number of tokens you can handle at a given time.
- If this method is not efficient, you can disable `HPUGraph`
completely. With HPU Graphs disabled, you are trading latency and
throughput at lower batches for potentially higher throughput on
higher batches. You can do that by adding `--enforce-eager` flag to
server (for online inference), or by passing `enforce_eager=True`
argument to LLM constructor (for offline inference).

402
docs/source/getting_started/gaudi-installation.rst
View File

@ -1,402 +0,0 @@
Installation with Intel® Gaudi® AI Accelerators
===============================================
This README provides instructions on running vLLM with Intel Gaudi devices.
Requirements and Installation
-----------------------------
Please follow the instructions provided in the `Gaudi Installation
Guide <https://docs.habana.ai/en/latest/Installation_Guide/index.html>`__
to set up the execution environment. To achieve the best performance,
please follow the methods outlined in the `Optimizing Training Platform
Guide <https://docs.habana.ai/en/latest/PyTorch/Model_Optimization_PyTorch/Optimization_in_Training_Platform.html>`__.
Requirements
~~~~~~~~~~~~
- OS: Ubuntu 22.04 LTS
- Python: 3.10
- Intel Gaudi accelerator
- Intel Gaudi software version 1.18.0
Quick start using Dockerfile
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
.. code:: console
$ docker build -f Dockerfile.hpu -t vllm-hpu-env .
$ docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --rm vllm-hpu-env
.. tip::
If you're observing the following error: ``docker: Error response from daemon: Unknown runtime specified habana.``, please refer to "Install Using Containers" section of `Intel Gaudi Software Stack and Driver Installation <https://docs.habana.ai/en/v1.18.0/Installation_Guide/Bare_Metal_Fresh_OS.html>`__. Make sure you have ``habana-container-runtime`` package installed and that ``habana`` container runtime is registered.
Build from source
~~~~~~~~~~~~~~~~~
Environment verification
^^^^^^^^^^^^^^^^^^^^^^^^
To verify that the Intel Gaudi software was correctly installed, run:
.. code:: console
$ hl-smi # verify that hl-smi is in your PATH and each Gaudi accelerator is visible
$ apt list --installed | grep habana # verify that habanalabs-firmware-tools, habanalabs-graph, habanalabs-rdma-core, habanalabs-thunk and habanalabs-container-runtime are installed
$ pip list | grep habana # verify that habana-torch-plugin, habana-torch-dataloader, habana-pyhlml and habana-media-loader are installed
$ pip list | grep neural # verify that neural_compressor is installed
Refer to `Intel Gaudi Software Stack
Verification <https://docs.habana.ai/en/latest/Installation_Guide/SW_Verification.html#platform-upgrade>`__
for more details.
Run Docker Image
^^^^^^^^^^^^^^^^
It is highly recommended to use the latest Docker image from Intel Gaudi
vault. Refer to the `Intel Gaudi
documentation <https://docs.habana.ai/en/latest/Installation_Guide/Bare_Metal_Fresh_OS.html#pull-prebuilt-containers>`__
for more details.
Use the following commands to run a Docker image:
.. code:: console
$ docker pull vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.0:latest
$ docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.18.0/ubuntu22.04/habanalabs/pytorch-installer-2.4.0:latest
Build and Install vLLM
^^^^^^^^^^^^^^^^^^^^^^
To build and install vLLM from source, run:
.. code:: console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ python setup.py develop
Currently, the latest features and performance optimizations are developed in Gaudi's `vLLM-fork <https://github.com/HabanaAI/vllm-fork>`__ and we periodically upstream them to vLLM main repo. To install latest `HabanaAI/vLLM-fork <https://github.com/HabanaAI/vllm-fork>`__, run the following:
.. code:: console
$ git clone https://github.com/HabanaAI/vllm-fork.git
$ cd vllm-fork
$ git checkout habana_main
$ python setup.py develop
Supported Features
------------------
- `Offline batched
inference <https://docs.vllm.ai/en/latest/getting_started/quickstart.html#offline-batched-inference>`__
- Online inference via `OpenAI-Compatible
Server <https://docs.vllm.ai/en/latest/getting_started/quickstart.html#openai-compatible-server>`__
- HPU autodetection - no need to manually select device within vLLM
- Paged KV cache with algorithms enabled for Intel Gaudi accelerators
- Custom Intel Gaudi implementations of Paged Attention, KV cache ops,
prefill attention, Root Mean Square Layer Normalization, Rotary
Positional Encoding
- Tensor parallelism support for multi-card inference
- Inference with `HPU Graphs <https://docs.habana.ai/en/latest/PyTorch/Inference_on_PyTorch/Inference_Using_HPU_Graphs.html>`__
for accelerating low-batch latency and throughput
- Attention with Linear Biases (ALiBi)
Unsupported Features
--------------------
- Beam search
- LoRA adapters
- Quantization
- Prefill chunking (mixed-batch inferencing)
Supported Configurations
------------------------
The following configurations have been validated to be function with
Gaudi2 devices. Configurations that are not listed may or may not work.
- `meta-llama/Llama-2-7b <https://huggingface.co/meta-llama/Llama-2-7b>`__
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- `meta-llama/Llama-2-7b-chat-hf <https://huggingface.co/meta-llama/Llama-2-7b-chat-hf>`__
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3-8B <https://huggingface.co/meta-llama/Meta-Llama-3-8B>`__
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3-8B-Instruct <https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct>`__
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3.1-8B <https://huggingface.co/meta-llama/Meta-Llama-3.1-8B>`__
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3.1-8B-Instruct <https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct>`__
on single HPU, or with tensor parallelism on 2x and 8x HPU, BF16
datatype with random or greedy sampling
- `meta-llama/Llama-2-70b <https://huggingface.co/meta-llama/Llama-2-70b>`__
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- `meta-llama/Llama-2-70b-chat-hf <https://huggingface.co/meta-llama/Llama-2-70b-chat-hf>`__
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3-70B <https://huggingface.co/meta-llama/Meta-Llama-3-70B>`__
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3-70B-Instruct <https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct>`__
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3.1-70B <https://huggingface.co/meta-llama/Meta-Llama-3.1-70B>`__
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
- `meta-llama/Meta-Llama-3.1-70B-Instruct <https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct>`__
with tensor parallelism on 8x HPU, BF16 datatype with random or greedy sampling
Performance Tuning
------------------
Execution modes
~~~~~~~~~~~~~~~
Currently in vLLM for HPU we support four execution modes, depending on selected HPU PyTorch Bridge backend (via ``PT_HPU_LAZY_MODE`` environment variable), and ``--enforce-eager`` flag.
.. list-table:: vLLM execution modes
:widths: 25 25 50
:header-rows: 1
* - ``PT_HPU_LAZY_MODE``
- ``enforce_eager``
- execution mode
* - 0
- 0
- torch.compile
* - 0
- 1
- PyTorch eager mode
* - 1
- 0
- HPU Graphs
* - 1
- 1
- PyTorch lazy mode
.. warning::
In 1.18.0, all modes utilizing ``PT_HPU_LAZY_MODE=0`` are highly experimental and should be only used for validating functional correctness. Their performance will be improved in the next releases. For obtaining the best performance in 1.18.0, please use HPU Graphs, or PyTorch lazy mode.
Bucketing mechanism
~~~~~~~~~~~~~~~~~~~
Intel Gaudi accelerators work best when operating on models with fixed tensor shapes. `Intel Gaudi Graph Compiler <https://docs.habana.ai/en/latest/Gaudi_Overview/Intel_Gaudi_Software_Suite.html#graph-compiler-and-runtime>`__ is responsible for generating optimized binary code that implements the given model topology on Gaudi. In its default configuration, the produced binary code may be heavily dependent on input and output tensor shapes, and can require graph recompilation when encountering differently shaped tensors within the same topology. While the resulting binaries utilize Gaudi efficiently, the compilation itself may introduce a noticeable overhead in end-to-end execution.
In a dynamic inference serving scenario, there is a need to minimize the number of graph compilations and reduce the risk of graph compilation occurring during server runtime. Currently it is achieved by "bucketing" model's forward pass across two dimensions - ``batch_size`` and ``sequence_length``.
.. note::
Bucketing allows us to reduce the number of required graphs significantly, but it does not handle any graph compilation and device code generation - this is done in warmup and HPUGraph capture phase.
Bucketing ranges are determined with 3 parameters - ``min``, ``step`` and ``max``. They can be set separately for prompt and decode phase, and for batch size and sequence length dimension. These parameters can be observed in logs during vLLM startup:
.. code-block::
INFO 08-01 21:37:59 hpu_model_runner.py:493] Prompt bucket config (min, step, max_warmup) bs:[1, 32, 4], seq:[128, 128, 1024]
INFO 08-01 21:37:59 hpu_model_runner.py:499] Generated 24 prompt buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024)]
INFO 08-01 21:37:59 hpu_model_runner.py:504] Decode bucket config (min, step, max_warmup) bs:[1, 128, 4], seq:[128, 128, 2048]
INFO 08-01 21:37:59 hpu_model_runner.py:509] Generated 48 decode buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (2, 1152), (2, 1280), (2, 1408), (2, 1536), (2, 1664), (2, 1792), (2, 1920), (2, 2048), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024), (4, 1152), (4, 1280), (4, 1408), (4, 1536), (4, 1664), (4, 1792), (4, 1920), (4, 2048)]
``min`` determines the lowest value of the bucket. ``step`` determines the interval between buckets, and ``max`` determines the upper bound of the bucket. Furthermore, interval between ``min`` and ``step`` has special handling - ``min`` gets multiplied by consecutive powers of two, until ``step`` gets reached. We call this the ramp-up phase and it is used for handling lower batch sizes with minimum wastage, while allowing larger padding on larger batch sizes.
Example (with ramp-up)
.. code-block::
min = 2, step = 32, max = 64
=> ramp_up = (2, 4, 8, 16)
=> stable = (32, 64)
=> buckets = ramp_up + stable => (2, 4, 8, 16, 32, 64)
Example (without ramp-up)
.. code-block::
min = 128, step = 128, max = 512
=> ramp_up = ()
=> stable = (128, 256, 384, 512)
=> buckets = ramp_up + stable => (128, 256, 384, 512)
In the logged scenario, 24 buckets were generated for prompt (prefill) runs, and 48 buckets for decode runs. Each bucket corresponds to a separate optimized device binary for a given model with specified tensor shapes. Whenever a batch of requests is processed, it is padded across batch and sequence length dimension to the smallest possible bucket.
.. warning::
If a request exceeds maximum bucket size in any dimension, it will be processed without padding, and its processing may require a graph compilation, potentially significantly increasing end-to-end latency. The boundaries of the buckets are user-configurable via environment variables, and upper bucket boundaries can be increased to avoid such scenario.
As an example, if a request of 3 sequences, with max sequence length of 412 comes in to an idle vLLM server, it will be padded executed as ``(4, 512)`` prefill bucket, as ``batch_size`` (number of sequences) will be padded to 4 (closest batch_size dimension higher than 3), and max sequence length will be padded to 512 (closest sequence length dimension higher than 412). After prefill stage, it will be executed as ``(4, 512)`` decode bucket and will continue as that bucket until either batch dimension changes (due to request being finished) - in which case it will become a ``(2, 512)`` bucket, or context length increases above 512 tokens, in which case it will become ``(4, 640)`` bucket.
.. note::
Bucketing is transparent to a client - padding in sequence length dimension is never returned to the client, and padding in batch dimension does not create new requests.
Warmup
~~~~~~
Warmup is an optional, but highly recommended step occurring before vLLM server starts listening. It executes a forward pass for each bucket with dummy data. The goal is to pre-compile all graphs and not incur any graph compilation overheads within bucket boundaries during server runtime. Each warmup step is logged during vLLM startup:
.. code-block::
INFO 08-01 22:26:47 hpu_model_runner.py:1066] [Warmup][Prompt][1/24] batch_size:4 seq_len:1024 free_mem:79.16 GiB
INFO 08-01 22:26:47 hpu_model_runner.py:1066] [Warmup][Prompt][2/24] batch_size:4 seq_len:896 free_mem:55.43 GiB
INFO 08-01 22:26:48 hpu_model_runner.py:1066] [Warmup][Prompt][3/24] batch_size:4 seq_len:768 free_mem:55.43 GiB
...
INFO 08-01 22:26:59 hpu_model_runner.py:1066] [Warmup][Prompt][24/24] batch_size:1 seq_len:128 free_mem:55.43 GiB
INFO 08-01 22:27:00 hpu_model_runner.py:1066] [Warmup][Decode][1/48] batch_size:4 seq_len:2048 free_mem:55.43 GiB
INFO 08-01 22:27:00 hpu_model_runner.py:1066] [Warmup][Decode][2/48] batch_size:4 seq_len:1920 free_mem:55.43 GiB
INFO 08-01 22:27:01 hpu_model_runner.py:1066] [Warmup][Decode][3/48] batch_size:4 seq_len:1792 free_mem:55.43 GiB
...
INFO 08-01 22:27:16 hpu_model_runner.py:1066] [Warmup][Decode][47/48] batch_size:2 seq_len:128 free_mem:55.43 GiB
INFO 08-01 22:27:16 hpu_model_runner.py:1066] [Warmup][Decode][48/48] batch_size:1 seq_len:128 free_mem:55.43 GiB
This example uses the same buckets as in *Bucketing mechanism* section. Each output line corresponds to execution of a single bucket. When bucket is executed for the first time, its graph is compiled and can be reused later on, skipping further graph compilations.
.. tip::
Compiling all the buckets might take some time and can be turned off with ``VLLM_SKIP_WARMUP=true`` environment variable. Keep in mind that if you do that, you may face graph compilations once executing a given bucket for the first time. It is fine to disable warmup for development, but it's highly recommended to enable it in deployment.
HPU Graph capture
~~~~~~~~~~~~~~~~~
`HPU Graphs <https://docs.habana.ai/en/latest/PyTorch/Inference_on_PyTorch/Inference_Using_HPU_Graphs.html>`__ are currently the most performant execution method of vLLM on Intel Gaudi. When HPU Graphs are enabled, execution graphs will be traced (recorded) ahead of time (after performing warmup), to be later replayed during inference, significantly reducing host overheads. Recording can take large amounts of memory, which needs to be taken into account when allocating KV cache. Enabling HPU Graphs will impact the number of available KV cache blocks, but vLLM provides user-configurable variables to control memory management.
When HPU Graphs are being used, they share the common memory pool ("usable memory") as KV cache, determined by ``gpu_memory_utilization`` flag (``0.9`` by default).
Before KV cache gets allocated, model weights are loaded onto the device, and a forward pass of the model is executed on dummy data, to estimate memory usage.
Only after that, ``gpu_memory_utilization`` flag is utilized - at its default value, will mark 90% of free device memory at that point as usable.
Next, KV cache gets allocated, model is warmed up, and HPU Graphs are captured.
Environment variable ``VLLM_GRAPH_RESERVED_MEM`` defines the ratio of memory reserved for HPU Graphs capture.
With its default value (``VLLM_GRAPH_RESERVED_MEM=0.1``), 10% of usable memory will be reserved for graph capture (later referred to as "usable graph memory"), and the remaining 90% will be utilized for KV cache.
Environment variable ``VLLM_GRAPH_PROMPT_RATIO`` determines the ratio of usable graph memory reserved for prefill and decode graphs. By default (``VLLM_GRAPH_PROMPT_RATIO=0.3``), both stages have equal memory constraints.
Lower value corresponds to less usable graph memory reserved for prefill stage, e.g. ``VLLM_GRAPH_PROMPT_RATIO=0.2`` will reserve 20% of usable graph memory for prefill graphs, and 80% of usable graph memory for decode graphs.
.. note::
``gpu_memory_utilization`` does not correspond to the absolute memory usage across HPU. It specifies the memory margin after loading the model and performing a profile run. If device has 100 GiB of total memory, and 50 GiB of free memory after loading model weights and executing profiling run, ``gpu_memory_utilization`` at its default value will mark 90% of 50 GiB as usable, leaving 5 GiB of margin, regardless of total device memory.
User can also configure the strategy for capturing HPU Graphs for prompt and decode stages separately. Strategy affects the order of capturing graphs. There are two strategies implemented:
- ``max_bs`` - graph capture queue will sorted in descending order by their batch sizes. Buckets with equal batch sizes are sorted by sequence length in ascending order (e.g. ``(64, 128)``, ``(64, 256)``, ``(32, 128)``, ``(32, 256)``, ``(1, 128)``, ``(1,256)``), default strategy for decode
- ``min_tokens`` - graph capture queue will be sorted in ascending order by the number of tokens each graph processes (``batch_size*sequence_length``), default strategy for prompt
When there's large amount of requests pending, vLLM scheduler will attempt to fill the maximum batch size for decode as soon as possible. When a request is finished, decode batch size decreases. When that happens, vLLM will attempt to schedule a prefill iteration for requests in the waiting queue, to fill the decode batch size to its previous state. This means that in a full load scenario, decode batch size is often at its maximum, which makes large batch size HPU Graphs crucial to capture, as reflected by ``max_bs`` strategy. On the other hand, prefills will be executed most frequently with very low batch sizes (1-4), which is reflected in ``min_tokens`` strategy.
.. note::
``VLLM_GRAPH_PROMPT_RATIO`` does not set a hard limit on memory taken by graphs for each stage (prefill and decode). vLLM will first attempt to use up entirety of usable prefill graph memory (usable graph memory * ``VLLM_GRAPH_PROMPT_RATIO``) for capturing prefill HPU Graphs, next it will attempt do the same for decode graphs and usable decode graph memory pool. If one stage is fully captured, and there is unused memory left within usable graph memory pool, vLLM will attempt further graph capture for the other stage, until no more HPU Graphs can be captured without exceeding reserved memory pool. The behavior on that mechanism can be observed in the example below.
Each described step is logged by vLLM server, as follows (negative values correspond to memory being released):
.. code-block::
INFO 08-02 17:37:44 hpu_model_runner.py:493] Prompt bucket config (min, step, max_warmup) bs:[1, 32, 4], seq:[128, 128, 1024]
INFO 08-02 17:37:44 hpu_model_runner.py:499] Generated 24 prompt buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024)]
INFO 08-02 17:37:44 hpu_model_runner.py:504] Decode bucket config (min, step, max_warmup) bs:[1, 128, 4], seq:[128, 128, 2048]
INFO 08-02 17:37:44 hpu_model_runner.py:509] Generated 48 decode buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (2, 1152), (2, 1280), (2, 1408), (2, 1536), (2, 1664), (2, 1792), (2, 1920), (2, 2048), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024), (4, 1152), (4, 1280), (4, 1408), (4, 1536), (4, 1664), (4, 1792), (4, 1920), (4, 2048)]
INFO 08-02 17:37:52 hpu_model_runner.py:430] Pre-loading model weights on hpu:0 took 14.97 GiB of device memory (14.97 GiB/94.62 GiB used) and 2.95 GiB of host memory (475.2 GiB/1007 GiB used)
INFO 08-02 17:37:52 hpu_model_runner.py:438] Wrapping in HPU Graph took 0 B of device memory (14.97 GiB/94.62 GiB used) and -252 KiB of host memory (475.2 GiB/1007 GiB used)
INFO 08-02 17:37:52 hpu_model_runner.py:442] Loading model weights took in total 14.97 GiB of device memory (14.97 GiB/94.62 GiB used) and 2.95 GiB of host memory (475.2 GiB/1007 GiB used)
INFO 08-02 17:37:54 hpu_worker.py:134] Model profiling run took 504 MiB of device memory (15.46 GiB/94.62 GiB used) and 180.9 MiB of host memory (475.4 GiB/1007 GiB used)
INFO 08-02 17:37:54 hpu_worker.py:158] Free device memory: 79.16 GiB, 39.58 GiB usable (gpu_memory_utilization=0.5), 15.83 GiB reserved for HPUGraphs (VLLM_GRAPH_RESERVED_MEM=0.4), 23.75 GiB reserved for KV cache
INFO 08-02 17:37:54 hpu_executor.py:85] # HPU blocks: 1519, # CPU blocks: 0
INFO 08-02 17:37:54 hpu_worker.py:190] Initializing cache engine took 23.73 GiB of device memory (39.2 GiB/94.62 GiB used) and -1.238 MiB of host memory (475.4 GiB/1007 GiB used)
INFO 08-02 17:37:54 hpu_model_runner.py:1066] [Warmup][Prompt][1/24] batch_size:4 seq_len:1024 free_mem:55.43 GiB
...
INFO 08-02 17:38:22 hpu_model_runner.py:1066] [Warmup][Decode][48/48] batch_size:1 seq_len:128 free_mem:55.43 GiB
INFO 08-02 17:38:22 hpu_model_runner.py:1159] Using 15.85 GiB/55.43 GiB of free device memory for HPUGraphs, 7.923 GiB for prompt and 7.923 GiB for decode (VLLM_GRAPH_PROMPT_RATIO=0.3)
INFO 08-02 17:38:22 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][1/24] batch_size:1 seq_len:128 free_mem:55.43 GiB
...
INFO 08-02 17:38:26 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][11/24] batch_size:1 seq_len:896 free_mem:48.77 GiB
INFO 08-02 17:38:27 hpu_model_runner.py:1066] [Warmup][Graph/Decode][1/48] batch_size:4 seq_len:128 free_mem:47.51 GiB
...
INFO 08-02 17:38:41 hpu_model_runner.py:1066] [Warmup][Graph/Decode][48/48] batch_size:1 seq_len:2048 free_mem:47.35 GiB
INFO 08-02 17:38:41 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][12/24] batch_size:4 seq_len:256 free_mem:47.35 GiB
INFO 08-02 17:38:42 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][13/24] batch_size:2 seq_len:512 free_mem:45.91 GiB
INFO 08-02 17:38:42 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][14/24] batch_size:1 seq_len:1024 free_mem:44.48 GiB
INFO 08-02 17:38:43 hpu_model_runner.py:1066] [Warmup][Graph/Prompt][15/24] batch_size:2 seq_len:640 free_mem:43.03 GiB
INFO 08-02 17:38:43 hpu_model_runner.py:1128] Graph/Prompt captured:15 (62.5%) used_mem:14.03 GiB buckets:[(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (4, 128), (4, 256)]
INFO 08-02 17:38:43 hpu_model_runner.py:1128] Graph/Decode captured:48 (100.0%) used_mem:161.9 MiB buckets:[(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (2, 1152), (2, 1280), (2, 1408), (2, 1536), (2, 1664), (2, 1792), (2, 1920), (2, 2048), (4, 128), (4, 256), (4, 384), (4, 512), (4, 640), (4, 768), (4, 896), (4, 1024), (4, 1152), (4, 1280), (4, 1408), (4, 1536), (4, 1664), (4, 1792), (4, 1920), (4, 2048)]
INFO 08-02 17:38:43 hpu_model_runner.py:1206] Warmup finished in 49 secs, allocated 14.19 GiB of device memory
INFO 08-02 17:38:43 hpu_executor.py:91] init_cache_engine took 37.92 GiB of device memory (53.39 GiB/94.62 GiB used) and 57.86 MiB of host memory (475.4 GiB/1007 GiB used)
Recommended vLLM Parameters
~~~~~~~~~~~~~~~~~~~~~~~~~~~
- We recommend running inference on Gaudi 2 with ``block_size`` of 128
for BF16 data type. Using default values (16, 32) might lead to
sub-optimal performance due to Matrix Multiplication Engine
under-utilization (see `Gaudi
Architecture <https://docs.habana.ai/en/latest/Gaudi_Overview/Gaudi_Architecture.html>`__).
- For max throughput on Llama 7B, we recommend running with batch size
of 128 or 256 and max context length of 2048 with HPU Graphs enabled.
If you encounter out-of-memory issues, see troubleshooting section.
Environment variables
~~~~~~~~~~~~~~~~~~~~~
**Diagnostic and profiling knobs:**
- ``VLLM_PROFILER_ENABLED``: if ``true``, high level profiler will be enabled. Resulting JSON traces can be viewed in `perfetto.habana.ai <https://perfetto.habana.ai/#!/viewer>`__. Disabled by default.
- ``VLLM_HPU_LOG_STEP_GRAPH_COMPILATION``: if ``true``, will log graph compilations per each vLLM engine step, only when there was any - highly recommended to use alongside ``PT_HPU_METRICS_GC_DETAILS=1``. Disabled by default.
- ``VLLM_HPU_LOG_STEP_GRAPH_COMPILATION_ALL``: if ``true``, will log graph compilations per each vLLM engine step, always, even if there were none. Disabled by default.
- ``VLLM_HPU_LOG_STEP_CPU_FALLBACKS``: if ``true``, will log cpu fallbacks per each vLLM engine step, only when there was any. Disabled by default.
- ``VLLM_HPU_LOG_STEP_CPU_FALLBACKS_ALL``: if ``true``, will log cpu fallbacks per each vLLM engine step, always, even if there were none. Disabled by default.
**Performance tuning knobs:**
- ``VLLM_SKIP_WARMUP``: if ``true``, warmup will be skipped, ``false`` by default
- ``VLLM_GRAPH_RESERVED_MEM``: percentage of memory dedicated for HPUGraph capture, ``0.1`` by default
- ``VLLM_GRAPH_PROMPT_RATIO``: percentage of reserved graph memory dedicated for prompt graphs, ``0.3`` by default
- ``VLLM_GRAPH_PROMPT_STRATEGY``: strategy determining order of prompt graph capture, ``min_tokens`` or ``max_bs``, ``min_tokens`` by default
- ``VLLM_GRAPH_DECODE_STRATEGY``: strategy determining order of decode graph capture, ``min_tokens`` or ``max_bs``, ``max_bs`` by default
- ``VLLM_{phase}_{dim}_BUCKET_{param}`` - collection of 12 environment variables configuring ranges of bucketing mechanism
- ``{phase}`` is either ``PROMPT`` or ``DECODE``
- ``{dim}`` is either ``BS``, ``SEQ`` or ``BLOCK``
- ``{param}`` is either ``MIN``, ``STEP`` or ``MAX``
- Default values:
- Prompt:
- batch size min (``VLLM_PROMPT_BS_BUCKET_MIN``): ``1``
- batch size step (``VLLM_PROMPT_BS_BUCKET_STEP``): ``min(max_num_seqs, 32)``
- batch size max (``VLLM_PROMPT_BS_BUCKET_MAX``): ``min(max_num_seqs, 64)``
- sequence length min (``VLLM_PROMPT_SEQ_BUCKET_MIN``): ``block_size``
- sequence length step (``VLLM_PROMPT_SEQ_BUCKET_STEP``): ``block_size``
- sequence length max (``VLLM_PROMPT_SEQ_BUCKET_MAX``): ``max_model_len``
- Decode:
- batch size min (``VLLM_DECODE_BS_BUCKET_MIN``): ``1``
- batch size step (``VLLM_DECODE_BS_BUCKET_STEP``): ``min(max_num_seqs, 32)``
- batch size max (``VLLM_DECODE_BS_BUCKET_MAX``): ``max_num_seqs``
- sequence length min (``VLLM_DECODE_BLOCK_BUCKET_MIN``): ``block_size``
- sequence length step (``VLLM_DECODE_BLOCK_BUCKET_STEP``): ``block_size``
- sequence length max (``VLLM_DECODE_BLOCK_BUCKET_MAX``): ``max(128, (max_num_seqs*max_model_len)/block_size)``
Additionally, there are HPU PyTorch Bridge environment variables impacting vLLM execution:
- ``PT_HPU_LAZY_MODE``: if ``0``, PyTorch Eager backend for Gaudi will be used, if ``1`` PyTorch Lazy backend for Gaudi will be used, ``1`` is default
- ``PT_HPU_ENABLE_LAZY_COLLECTIVES``: required to be ``true`` for tensor parallel inference with HPU Graphs
Troubleshooting: Tweaking HPU Graphs
------------------------------------
If you experience device out-of-memory issues or want to attempt
inference at higher batch sizes, try tweaking HPU Graphs by following
the below:
- Tweak ``gpu_memory_utilization`` knob. It will decrease the
allocation of KV cache, leaving some headroom for capturing graphs
with larger batch size. By default ``gpu_memory_utilization`` is set
to 0.9. It attempts to allocate ~90% of HBM left for KV cache after
short profiling run. Note that decreasing reduces the number of KV
cache blocks you have available, and therefore reduces the effective
maximum number of tokens you can handle at a given time.
- If this method is not efficient, you can disable ``HPUGraph``
completely. With HPU Graphs disabled, you are trading latency and
throughput at lower batches for potentially higher throughput on
higher batches. You can do that by adding ``--enforce-eager`` flag to
server (for online inference), or by passing ``enforce_eager=True``
argument to LLM constructor (for offline inference).

199
docs/source/getting_started/installation.md Normal file
View File

@ -0,0 +1,199 @@
(installation)=
# Installation
vLLM is a Python library that also contains pre-compiled C++ and CUDA (12.1) binaries.
## Requirements
- OS: Linux
- Python: 3.9 -- 3.12
- GPU: compute capability 7.0 or higher (e.g., V100, T4, RTX20xx, A100, L4, H100, etc.)
## Install released versions
You can install vLLM using pip:
```console
$ # (Recommended) Create a new conda environment.
$ conda create -n myenv python=3.12 -y
$ conda activate myenv
$ # Install vLLM with CUDA 12.1.
$ pip install vllm
```
```{note}
Although we recommend using `conda` to create and manage Python environments, it is highly recommended to use `pip` to install vLLM. This is because `pip` can install `torch` with separate library packages like `NCCL`, while `conda` installs `torch` with statically linked `NCCL`. This can cause issues when vLLM tries to use `NCCL`. See [this issue](https://github.com/vllm-project/vllm/issues/8420) for more details.
```
````{note}
As of now, vLLM's binaries are compiled with CUDA 12.1 and public PyTorch release versions by default.
We also provide vLLM binaries compiled with CUDA 11.8 and public PyTorch release versions:
```console
$ # Install vLLM with CUDA 11.8.
$ export VLLM_VERSION=0.6.1.post1
$ export PYTHON_VERSION=310
$ pip install https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cu118-cp${PYTHON_VERSION}-cp${PYTHON_VERSION}-manylinux1_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cu118
```
In order to be performant, vLLM has to compile many cuda kernels. The compilation unfortunately introduces binary incompatibility with other CUDA versions and PyTorch versions, even for the same PyTorch version with different building configurations.
Therefore, it is recommended to install vLLM with a **fresh new** conda environment. If either you have a different CUDA version or you want to use an existing PyTorch installation, you need to build vLLM from source. See below for instructions.
````
(install-the-latest-code)=
## Install the latest code
LLM inference is a fast-evolving field, and the latest code may contain bug fixes, performance improvements, and new features that are not released yet. To allow users to try the latest code without waiting for the next release, vLLM provides wheels for Linux running on a x86 platform with CUDA 12 for every commit since `v0.5.3`. You can download and install it with the following command:
```console
$ pip install https://vllm-wheels.s3.us-west-2.amazonaws.com/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
```
If you want to access the wheels for previous commits, you can specify the commit hash in the URL:
```console
$ export VLLM_COMMIT=33f460b17a54acb3b6cc0b03f4a17876cff5eafd # use full commit hash from the main branch
$ pip install https://vllm-wheels.s3.us-west-2.amazonaws.com/${VLLM_COMMIT}/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
```
Note that the wheels are built with Python 3.8 ABI (see [PEP 425](https://peps.python.org/pep-0425/) for more details about ABI), so **they are compatible with Python 3.8 and later**. The version string in the wheel file name (`1.0.0.dev`) is just a placeholder to have a unified URL for the wheels. The actual versions of wheels are contained in the wheel metadata. Although we don't support Python 3.8 any more (because PyTorch 2.5 dropped support for Python 3.8), the wheels are still built with Python 3.8 ABI to keep the same wheel name as before.
Another way to access the latest code is to use the docker images:
```console
$ export VLLM_COMMIT=33f460b17a54acb3b6cc0b03f4a17876cff5eafd # use full commit hash from the main branch
$ docker pull public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:${VLLM_COMMIT}
```
These docker images are used for CI and testing only, and they are not intended for production use. They will be expired after several days.
The latest code can contain bugs and may not be stable. Please use it with caution.
(build-from-source)=
## Build from source
(python-only-build)=
### Python-only build (without compilation)
If you only need to change Python code, you can build and install vLLM without compilation. Using `pip`'s [`--editable` flag](https://pip.pypa.io/en/stable/topics/local-project-installs/#editable-installs), changes you make to the code will be reflected when you run vLLM:
```console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ VLLM_USE_PRECOMPILED=1 pip install --editable .
```
This will download the latest nightly wheel and use the compiled libraries from there in the install.
The `VLLM_PRECOMPILED_WHEEL_LOCATION` environment variable can be used instead of `VLLM_USE_PRECOMPILED` to specify a custom path or URL to the wheel file. For example, to use the [0.6.1.post1 PyPi wheel](https://pypi.org/project/vllm/#files):
```console
$ export VLLM_PRECOMPILED_WHEEL_LOCATION=https://files.pythonhosted.org/packages/4a/4c/ee65ba33467a4c0de350ce29fbae39b9d0e7fcd887cc756fa993654d1228/vllm-0.6.3.post1-cp38-abi3-manylinux1_x86_64.whl
$ pip install --editable .
```
You can find more information about vLLM's wheels [above](#install-the-latest-code).
```{note}
There is a possibility that your source code may have a different commit ID compared to the latest vLLM wheel, which could potentially lead to unknown errors.
It is recommended to use the same commit ID for the source code as the vLLM wheel you have installed. Please refer to [the section above](#install-the-latest-code) for instructions on how to install a specified wheel.
```
### Full build (with compilation)
If you want to modify C++ or CUDA code, you'll need to build vLLM from source. This can take several minutes:
```console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ pip install -e .
```
```{tip}
Building from source requires a lot of compilation. If you are building from source repeatedly, it's more efficient to cache the compilation results.
For example, you can install [ccache](https://github.com/ccache/ccache) using `conda install ccache` or `apt install ccache` .
As long as `which ccache` command can find the `ccache` binary, it will be used automatically by the build system. After the first build, subsequent builds will be much faster.
[sccache](https://github.com/mozilla/sccache) works similarly to `ccache`, but has the capability to utilize caching in remote storage environments.
The following environment variables can be set to configure the vLLM `sccache` remote: `SCCACHE_BUCKET=vllm-build-sccache SCCACHE_REGION=us-west-2 SCCACHE_S3_NO_CREDENTIALS=1`. We also recommend setting `SCCACHE_IDLE_TIMEOUT=0`.
```
#### Use an existing PyTorch installation
There are scenarios where the PyTorch dependency cannot be easily installed via pip, e.g.:
- Building vLLM with PyTorch nightly or a custom PyTorch build.
- Building vLLM with aarch64 and CUDA (GH200), where the PyTorch wheels are not available on PyPI. Currently, only the PyTorch nightly has wheels for aarch64 with CUDA. You can run `pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu124` to [install PyTorch nightly](https://pytorch.org/get-started/locally/), and then build vLLM on top of it.
To build vLLM using an existing PyTorch installation:
```console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ python use_existing_torch.py
$ pip install -r requirements-build.txt
$ pip install -e . --no-build-isolation
```
#### Use the local cutlass for compilation
Currently, before starting the build process, vLLM fetches cutlass code from GitHub. However, there may be scenarios where you want to use a local version of cutlass instead.
To achieve this, you can set the environment variable VLLM_CUTLASS_SRC_DIR to point to your local cutlass directory.
```console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ VLLM_CUTLASS_SRC_DIR=/path/to/cutlass pip install -e .
```
#### Troubleshooting
To avoid your system being overloaded, you can limit the number of compilation jobs
to be run simultaneously, via the environment variable `MAX_JOBS`. For example:
```console
$ export MAX_JOBS=6
$ pip install -e .
```
This is especially useful when you are building on less powerful machines. For example, when you use WSL it only [assigns 50% of the total memory by default](https://learn.microsoft.com/en-us/windows/wsl/wsl-config#main-wsl-settings), so using `export MAX_JOBS=1` can avoid compiling multiple files simultaneously and running out of memory.
A side effect is a much slower build process.
Additionally, if you have trouble building vLLM, we recommend using the NVIDIA PyTorch Docker image.
```console
$ # Use `--ipc=host` to make sure the shared memory is large enough.
$ docker run --gpus all -it --rm --ipc=host nvcr.io/nvidia/pytorch:23.10-py3
```
If you don't want to use docker, it is recommended to have a full installation of CUDA Toolkit. You can download and install it from [the official website](https://developer.nvidia.com/cuda-toolkit-archive). After installation, set the environment variable `CUDA_HOME` to the installation path of CUDA Toolkit, and make sure that the `nvcc` compiler is in your `PATH`, e.g.:
```console
$ export CUDA_HOME=/usr/local/cuda
$ export PATH="${CUDA_HOME}/bin:$PATH"
```
Here is a sanity check to verify that the CUDA Toolkit is correctly installed:
```console
$ nvcc --version # verify that nvcc is in your PATH
$ ${CUDA_HOME}/bin/nvcc --version # verify that nvcc is in your CUDA_HOME
```
### Unsupported OS build
vLLM can fully run only on Linux but for development purposes, you can still build it on other systems (for example, macOS), allowing for imports and a more convenient development environment. The binaries will not be compiled and won't work on non-Linux systems.
Simply disable the `VLLM_TARGET_DEVICE` environment variable before installing:
```console
$ export VLLM_TARGET_DEVICE=empty
$ pip install -e .
```

214
docs/source/getting_started/installation.rst
View File

@ -1,214 +0,0 @@
.. _installation:
============
Installation
============
vLLM is a Python library that also contains pre-compiled C++ and CUDA (12.1) binaries.
Requirements
============
* OS: Linux
* Python: 3.9 -- 3.12
* GPU: compute capability 7.0 or higher (e.g., V100, T4, RTX20xx, A100, L4, H100, etc.)
Install released versions
=========================
You can install vLLM using pip:
.. code-block:: console
$ # (Recommended) Create a new conda environment.
$ conda create -n myenv python=3.12 -y
$ conda activate myenv
$ # Install vLLM with CUDA 12.1.
$ pip install vllm
.. note::
Although we recommend using ``conda`` to create and manage Python environments, it is highly recommended to use ``pip`` to install vLLM. This is because ``pip`` can install ``torch`` with separate library packages like ``NCCL``, while ``conda`` installs ``torch`` with statically linked ``NCCL``. This can cause issues when vLLM tries to use ``NCCL``. See `this issue <https://github.com/vllm-project/vllm/issues/8420>`_ for more details.
.. note::
As of now, vLLM's binaries are compiled with CUDA 12.1 and public PyTorch release versions by default.
We also provide vLLM binaries compiled with CUDA 11.8 and public PyTorch release versions:
.. code-block:: console
$ # Install vLLM with CUDA 11.8.
$ export VLLM_VERSION=0.6.1.post1
$ export PYTHON_VERSION=310
$ pip install https://github.com/vllm-project/vllm/releases/download/v${VLLM_VERSION}/vllm-${VLLM_VERSION}+cu118-cp${PYTHON_VERSION}-cp${PYTHON_VERSION}-manylinux1_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cu118
In order to be performant, vLLM has to compile many cuda kernels. The compilation unfortunately introduces binary incompatibility with other CUDA versions and PyTorch versions, even for the same PyTorch version with different building configurations.
Therefore, it is recommended to install vLLM with a **fresh new** conda environment. If either you have a different CUDA version or you want to use an existing PyTorch installation, you need to build vLLM from source. See below for instructions.
.. _install-the-latest-code:
Install the latest code
=======================
LLM inference is a fast-evolving field, and the latest code may contain bug fixes, performance improvements, and new features that are not released yet. To allow users to try the latest code without waiting for the next release, vLLM provides wheels for Linux running on a x86 platform with CUDA 12 for every commit since ``v0.5.3``. You can download and install it with the following command:
.. code-block:: console
$ pip install https://vllm-wheels.s3.us-west-2.amazonaws.com/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
If you want to access the wheels for previous commits, you can specify the commit hash in the URL:
.. code-block:: console
$ export VLLM_COMMIT=33f460b17a54acb3b6cc0b03f4a17876cff5eafd # use full commit hash from the main branch
$ pip install https://vllm-wheels.s3.us-west-2.amazonaws.com/${VLLM_COMMIT}/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl
Note that the wheels are built with Python 3.8 ABI (see `PEP 425 <https://peps.python.org/pep-0425/>`_ for more details about ABI), so **they are compatible with Python 3.8 and later**. The version string in the wheel file name (``1.0.0.dev``) is just a placeholder to have a unified URL for the wheels. The actual versions of wheels are contained in the wheel metadata. Although we don't support Python 3.8 any more (because PyTorch 2.5 dropped support for Python 3.8), the wheels are still built with Python 3.8 ABI to keep the same wheel name as before.
Another way to access the latest code is to use the docker images:
.. code-block:: console
$ export VLLM_COMMIT=33f460b17a54acb3b6cc0b03f4a17876cff5eafd # use full commit hash from the main branch
$ docker pull public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:${VLLM_COMMIT}
These docker images are used for CI and testing only, and they are not intended for production use. They will be expired after several days.
The latest code can contain bugs and may not be stable. Please use it with caution.
.. _build_from_source:
Build from source
=================
.. _python-only-build:
Python-only build (without compilation)
---------------------------------------
If you only need to change Python code, you can build and install vLLM without compilation. Using `pip's ``--editable`` flag <https://pip.pypa.io/en/stable/topics/local-project-installs/#editable-installs>`_, changes you make to the code will be reflected when you run vLLM:
.. code-block:: console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ VLLM_USE_PRECOMPILED=1 pip install --editable .
This will download the latest nightly wheel and use the compiled libraries from there in the install.
The ``VLLM_PRECOMPILED_WHEEL_LOCATION`` environment variable can be used instead of ``VLLM_USE_PRECOMPILED`` to specify a custom path or URL to the wheel file. For example, to use the `0.6.1.post1 PyPi wheel <https://pypi.org/project/vllm/#files>`_:
.. code-block:: console
$ export VLLM_PRECOMPILED_WHEEL_LOCATION=https://files.pythonhosted.org/packages/4a/4c/ee65ba33467a4c0de350ce29fbae39b9d0e7fcd887cc756fa993654d1228/vllm-0.6.3.post1-cp38-abi3-manylinux1_x86_64.whl
$ pip install --editable .
You can find more information about vLLM's wheels `above <#install-the-latest-code>`_.
.. note::
There is a possibility that your source code may have a different commit ID compared to the latest vLLM wheel, which could potentially lead to unknown errors.
It is recommended to use the same commit ID for the source code as the vLLM wheel you have installed. Please refer to `the section above <#install-the-latest-code>`_ for instructions on how to install a specified wheel.
Full build (with compilation)
-----------------------------
If you want to modify C++ or CUDA code, you'll need to build vLLM from source. This can take several minutes:
.. code-block:: console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ pip install -e .
.. tip::
Building from source requires a lot of compilation. If you are building from source repeatedly, it's more efficient to cache the compilation results.
For example, you can install `ccache <https://github.com/ccache/ccache>`_ using ``conda install ccache`` or ``apt install ccache`` .
As long as ``which ccache`` command can find the ``ccache`` binary, it will be used automatically by the build system. After the first build, subsequent builds will be much faster.
`sccache <https://github.com/mozilla/sccache>`_ works similarly to ``ccache``, but has the capability to utilize caching in remote storage environments.
The following environment variables can be set to configure the vLLM ``sccache`` remote: ``SCCACHE_BUCKET=vllm-build-sccache SCCACHE_REGION=us-west-2 SCCACHE_S3_NO_CREDENTIALS=1``. We also recommend setting ``SCCACHE_IDLE_TIMEOUT=0``.
Use an existing PyTorch installation
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
There are scenarios where the PyTorch dependency cannot be easily installed via pip, e.g.:
* Building vLLM with PyTorch nightly or a custom PyTorch build.
* Building vLLM with aarch64 and CUDA (GH200), where the PyTorch wheels are not available on PyPI. Currently, only the PyTorch nightly has wheels for aarch64 with CUDA. You can run ``pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu124`` to `install PyTorch nightly <https://pytorch.org/get-started/locally/>`_, and then build vLLM on top of it.
To build vLLM using an existing PyTorch installation:
.. code-block:: console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ python use_existing_torch.py
$ pip install -r requirements-build.txt
$ pip install -e . --no-build-isolation
Use the local cutlass for compilation
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Currently, before starting the build process, vLLM fetches cutlass code from GitHub. However, there may be scenarios where you want to use a local version of cutlass instead.
To achieve this, you can set the environment variable VLLM_CUTLASS_SRC_DIR to point to your local cutlass directory.
.. code-block:: console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ VLLM_CUTLASS_SRC_DIR=/path/to/cutlass pip install -e .
Troubleshooting
~~~~~~~~~~~~~~~
To avoid your system being overloaded, you can limit the number of compilation jobs
to be run simultaneously, via the environment variable ``MAX_JOBS``. For example:
.. code-block:: console
$ export MAX_JOBS=6
$ pip install -e .
This is especially useful when you are building on less powerful machines. For example, when you use WSL it only `assigns 50% of the total memory by default <https://learn.microsoft.com/en-us/windows/wsl/wsl-config#main-wsl-settings>`_, so using ``export MAX_JOBS=1`` can avoid compiling multiple files simultaneously and running out of memory.
A side effect is a much slower build process.
Additionally, if you have trouble building vLLM, we recommend using the NVIDIA PyTorch Docker image.
.. code-block:: console
$ # Use `--ipc=host` to make sure the shared memory is large enough.
$ docker run --gpus all -it --rm --ipc=host nvcr.io/nvidia/pytorch:23.10-py3
If you don't want to use docker, it is recommended to have a full installation of CUDA Toolkit. You can download and install it from `the official website <https://developer.nvidia.com/cuda-toolkit-archive>`_. After installation, set the environment variable ``CUDA_HOME`` to the installation path of CUDA Toolkit, and make sure that the ``nvcc`` compiler is in your ``PATH``, e.g.:
.. code-block:: console
$ export CUDA_HOME=/usr/local/cuda
$ export PATH="${CUDA_HOME}/bin:$PATH"
Here is a sanity check to verify that the CUDA Toolkit is correctly installed:
.. code-block:: console
$ nvcc --version # verify that nvcc is in your PATH
$ ${CUDA_HOME}/bin/nvcc --version # verify that nvcc is in your CUDA_HOME
Unsupported OS build
--------------------
vLLM can fully run only on Linux but for development purposes, you can still build it on other systems (for example, macOS), allowing for imports and a more convenient development environment. The binaries will not be compiled and won't work on non-Linux systems.
Simply disable the ``VLLM_TARGET_DEVICE`` environment variable before installing:
.. code-block:: console
$ export VLLM_TARGET_DEVICE=empty
$ pip install -e .

132
docs/source/getting_started/neuron-installation.md Normal file
View File

@ -0,0 +1,132 @@
(installation-neuron)=
# Installation with Neuron
vLLM 0.3.3 onwards supports model inferencing and serving on AWS Trainium/Inferentia with Neuron SDK with continuous batching.
Paged Attention and Chunked Prefill are currently in development and will be available soon.
Data types currently supported in Neuron SDK are FP16 and BF16.
## Requirements
- OS: Linux
- Python: 3.9 -- 3.11
- Accelerator: NeuronCore_v2 (in trn1/inf2 instances)
- Pytorch 2.0.1/2.1.1
- AWS Neuron SDK 2.16/2.17 (Verified on python 3.8)
Installation steps:
- [Build from source](#build-from-source-neuron)
- [Step 0. Launch Trn1/Inf2 instances](#launch-instances)
- [Step 1. Install drivers and tools](#install-drivers)
- [Step 2. Install transformers-neuronx and its dependencies](#install-tnx)
- [Step 3. Install vLLM from source](#install-vllm)
(build-from-source-neuron)=
```{note}
The currently supported version of Pytorch for Neuron installs `triton` version `2.1.0`. This is incompatible with vLLM >= 0.5.3. You may see an error `cannot import name 'default_dump_dir...`. To work around this, run a `pip install --upgrade triton==3.0.0` after installing the vLLM wheel.
```
## Build from source
Following instructions are applicable to Neuron SDK 2.16 and beyond.
(launch-instances)=
### Step 0. Launch Trn1/Inf2 instances
Here are the steps to launch trn1/inf2 instances, in order to install [PyTorch Neuron ("torch-neuronx") Setup on Ubuntu 22.04 LTS](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/general/setup/neuron-setup/pytorch/neuronx/ubuntu/torch-neuronx-ubuntu22.html).
- Please follow the instructions at [launch an Amazon EC2 Instance](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EC2_GetStarted.html#ec2-launch-instance) to launch an instance. When choosing the instance type at the EC2 console, please make sure to select the correct instance type.
- To get more information about instances sizes and pricing see: [Trn1 web page](https://aws.amazon.com/ec2/instance-types/trn1/), [Inf2 web page](https://aws.amazon.com/ec2/instance-types/inf2/)
- Select Ubuntu Server 22.04 TLS AMI
- When launching a Trn1/Inf2, please adjust your primary EBS volume size to a minimum of 512GB.
- After launching the instance, follow the instructions in [Connect to your instance](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/AccessingInstancesLinux.html) to connect to the instance
(install-drivers)=
### Step 1. Install drivers and tools
The installation of drivers and tools wouldn't be necessary, if [Deep Learning AMI Neuron](https://docs.aws.amazon.com/dlami/latest/devguide/appendix-ami-release-notes.html) is installed. In case the drivers and tools are not installed on the operating system, follow the steps below:
```console
# Configure Linux for Neuron repository updates
. /etc/os-release
sudo tee /etc/apt/sources.list.d/neuron.list > /dev/null <<EOF
deb https://apt.repos.neuron.amazonaws.com ${VERSION_CODENAME} main
EOF
wget -qO - https://apt.repos.neuron.amazonaws.com/GPG-PUB-KEY-AMAZON-AWS-NEURON.PUB | sudo apt-key add -
# Update OS packages
sudo apt-get update -y
# Install OS headers
sudo apt-get install linux-headers-$(uname -r) -y
# Install git
sudo apt-get install git -y
# install Neuron Driver
sudo apt-get install aws-neuronx-dkms=2.* -y
# Install Neuron Runtime
sudo apt-get install aws-neuronx-collectives=2.* -y
sudo apt-get install aws-neuronx-runtime-lib=2.* -y
# Install Neuron Tools
sudo apt-get install aws-neuronx-tools=2.* -y
# Add PATH
export PATH=/opt/aws/neuron/bin:$PATH
```
(install-tnx)=
### Step 2. Install transformers-neuronx and its dependencies
[transformers-neuronx](https://github.com/aws-neuron/transformers-neuronx) will be the backend to support inference on trn1/inf2 instances.
Follow the steps below to install transformer-neuronx package and its dependencies.
```console
# Install Python venv
sudo apt-get install -y python3.10-venv g++
# Create Python venv
python3.10 -m venv aws_neuron_venv_pytorch
# Activate Python venv
source aws_neuron_venv_pytorch/bin/activate
# Install Jupyter notebook kernel
pip install ipykernel
python3.10 -m ipykernel install --user --name aws_neuron_venv_pytorch --display-name "Python (torch-neuronx)"
pip install jupyter notebook
pip install environment_kernels
# Set pip repository pointing to the Neuron repository
python -m pip config set global.extra-index-url https://pip.repos.neuron.amazonaws.com
# Install wget, awscli
python -m pip install wget
python -m pip install awscli
# Update Neuron Compiler and Framework
python -m pip install --upgrade neuronx-cc==2.* --pre torch-neuronx==2.1.* torchvision transformers-neuronx
```
(install-vllm)=
### Step 3. Install vLLM from source
Once neuronx-cc and transformers-neuronx packages are installed, we will be able to install vllm as follows:
```console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ pip install -U -r requirements-neuron.txt
$ VLLM_TARGET_DEVICE="neuron" pip install .
```
If neuron packages are detected correctly in the installation process, `vllm-0.3.0+neuron212` will be installed.

140
docs/source/getting_started/neuron-installation.rst
View File

@ -1,140 +0,0 @@
.. _installation_neuron:
Installation with Neuron
========================
vLLM 0.3.3 onwards supports model inferencing and serving on AWS Trainium/Inferentia with Neuron SDK with continuous batching.
Paged Attention and Chunked Prefill are currently in development and will be available soon.
Data types currently supported in Neuron SDK are FP16 and BF16.
Requirements
------------
* OS: Linux
* Python: 3.9 -- 3.11
* Accelerator: NeuronCore_v2 (in trn1/inf2 instances)
* Pytorch 2.0.1/2.1.1
* AWS Neuron SDK 2.16/2.17 (Verified on python 3.8)
Installation steps:
- :ref:`Build from source <build_from_source_neuron>`
- :ref:`Step 0. Launch Trn1/Inf2 instances <launch_instances>`
- :ref:`Step 1. Install drivers and tools <install_drivers>`
- :ref:`Step 2. Install transformers-neuronx and its dependencies <install_tnx>`
- :ref:`Step 3. Install vLLM from source <install_vllm>`
.. _build_from_source_neuron:
.. note::
The currently supported version of Pytorch for Neuron installs `triton` version `2.1.0`. This is incompatible with vLLM >= 0.5.3. You may see an error `cannot import name 'default_dump_dir...`. To work around this, run a `pip install --upgrade triton==3.0.0` after installing the vLLM wheel.
Build from source
-----------------
Following instructions are applicable to Neuron SDK 2.16 and beyond.
.. _launch_instances:
Step 0. Launch Trn1/Inf2 instances
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Here are the steps to launch trn1/inf2 instances, in order to install `PyTorch Neuron ("torch-neuronx") Setup on Ubuntu 22.04 LTS <https://awsdocs-neuron.readthedocs-hosted.com/en/latest/general/setup/neuron-setup/pytorch/neuronx/ubuntu/torch-neuronx-ubuntu22.html>`_.
- Please follow the instructions at `launch an Amazon EC2 Instance <https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EC2_GetStarted.html#ec2-launch-instance>`_ to launch an instance. When choosing the instance type at the EC2 console, please make sure to select the correct instance type.
- To get more information about instances sizes and pricing see: `Trn1 web page <https://aws.amazon.com/ec2/instance-types/trn1/>`_, `Inf2 web page <https://aws.amazon.com/ec2/instance-types/inf2/>`_
- Select Ubuntu Server 22.04 TLS AMI
- When launching a Trn1/Inf2, please adjust your primary EBS volume size to a minimum of 512GB.
- After launching the instance, follow the instructions in `Connect to your instance <https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/AccessingInstancesLinux.html>`_ to connect to the instance
.. _install_drivers:
Step 1. Install drivers and tools
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The installation of drivers and tools wouldn't be necessary, if `Deep Learning AMI Neuron <https://docs.aws.amazon.com/dlami/latest/devguide/appendix-ami-release-notes.html>`_ is installed. In case the drivers and tools are not installed on the operating system, follow the steps below:
.. code-block:: console
# Configure Linux for Neuron repository updates
. /etc/os-release
sudo tee /etc/apt/sources.list.d/neuron.list > /dev/null <<EOF
deb https://apt.repos.neuron.amazonaws.com ${VERSION_CODENAME} main
EOF
wget -qO - https://apt.repos.neuron.amazonaws.com/GPG-PUB-KEY-AMAZON-AWS-NEURON.PUB | sudo apt-key add -
# Update OS packages
sudo apt-get update -y
# Install OS headers
sudo apt-get install linux-headers-$(uname -r) -y
# Install git
sudo apt-get install git -y
# install Neuron Driver
sudo apt-get install aws-neuronx-dkms=2.* -y
# Install Neuron Runtime
sudo apt-get install aws-neuronx-collectives=2.* -y
sudo apt-get install aws-neuronx-runtime-lib=2.* -y
# Install Neuron Tools
sudo apt-get install aws-neuronx-tools=2.* -y
# Add PATH
export PATH=/opt/aws/neuron/bin:$PATH
.. _install_tnx:
Step 2. Install transformers-neuronx and its dependencies
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
`transformers-neuronx <https://github.com/aws-neuron/transformers-neuronx>`_ will be the backend to support inference on trn1/inf2 instances.
Follow the steps below to install transformer-neuronx package and its dependencies.
.. code-block:: console
# Install Python venv
sudo apt-get install -y python3.10-venv g++
# Create Python venv
python3.10 -m venv aws_neuron_venv_pytorch
# Activate Python venv
source aws_neuron_venv_pytorch/bin/activate
# Install Jupyter notebook kernel
pip install ipykernel
python3.10 -m ipykernel install --user --name aws_neuron_venv_pytorch --display-name "Python (torch-neuronx)"
pip install jupyter notebook
pip install environment_kernels
# Set pip repository pointing to the Neuron repository
python -m pip config set global.extra-index-url https://pip.repos.neuron.amazonaws.com
# Install wget, awscli
python -m pip install wget
python -m pip install awscli
# Update Neuron Compiler and Framework
python -m pip install --upgrade neuronx-cc==2.* --pre torch-neuronx==2.1.* torchvision transformers-neuronx
.. _install_vllm:
Step 3. Install vLLM from source
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Once neuronx-cc and transformers-neuronx packages are installed, we will be able to install vllm as follows:
.. code-block:: console
$ git clone https://github.com/vllm-project/vllm.git
$ cd vllm
$ pip install -U -r requirements-neuron.txt
$ VLLM_TARGET_DEVICE="neuron" pip install .
If neuron packages are detected correctly in the installation process, ``vllm-0.3.0+neuron212`` will be installed.

104
docs/source/getting_started/openvino-installation.md Normal file
View File

@ -0,0 +1,104 @@
(installation-openvino)=
# Installation with OpenVINO
vLLM powered by OpenVINO supports all LLM models from {doc}`vLLM supported models list <../models/supported_models>` and can perform optimal model serving on all x86-64 CPUs with, at least, AVX2 support, as well as on both integrated and discrete Intel® GPUs ([the list of supported GPUs](https://docs.openvino.ai/2024/about-openvino/release-notes-openvino/system-requirements.html#gpu)). OpenVINO vLLM backend supports the following advanced vLLM features:
- Prefix caching (`--enable-prefix-caching`)
- Chunked prefill (`--enable-chunked-prefill`)
**Table of contents**:
- [Requirements](#openvino-backend-requirements)
- [Quick start using Dockerfile](#openvino-backend-quick-start-dockerfile)
- [Build from source](#install-openvino-backend-from-source)
- [Performance tips](#openvino-backend-performance-tips)
- [Limitations](#openvino-backend-limitations)
(openvino-backend-requirements)=
## Requirements
- OS: Linux
- Instruction set architecture (ISA) requirement: at least AVX2.
(openvino-backend-quick-start-dockerfile)=
## Quick start using Dockerfile
```console
$ docker build -f Dockerfile.openvino -t vllm-openvino-env .
$ docker run -it --rm vllm-openvino-env
```
(install-openvino-backend-from-source)=
## Install from source
- First, install Python. For example, on Ubuntu 22.04, you can run:
```console
$ sudo apt-get update -y
$ sudo apt-get install python3
```
- Second, install prerequisites vLLM OpenVINO backend installation:
```console
$ pip install --upgrade pip
$ pip install -r requirements-build.txt --extra-index-url https://download.pytorch.org/whl/cpu
```
- Finally, install vLLM with OpenVINO backend:
```console
$ PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" VLLM_TARGET_DEVICE=openvino python -m pip install -v .
```
- [Optional] To use vLLM OpenVINO backend with a GPU device, ensure your system is properly set up. Follow the instructions provided here: [https://docs.openvino.ai/2024/get-started/configurations/configurations-intel-gpu.html](https://docs.openvino.ai/2024/get-started/configurations/configurations-intel-gpu.html).
(openvino-backend-performance-tips)=
## Performance tips
### vLLM OpenVINO backend environment variables
- `VLLM_OPENVINO_DEVICE` to specify which device utilize for the inference. If there are multiple GPUs in the system, additional indexes can be used to choose the proper one (e.g, `VLLM_OPENVINO_DEVICE=GPU.1`). If the value is not specified, CPU device is used by default.
- `VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS=ON` to enable U8 weights compression during model loading stage. By default, compression is turned off. You can also export model with different compression techniques using `optimum-cli` and pass exported folder as `<model_id>`
### CPU performance tips
CPU uses the following environment variables to control behavior:
- `VLLM_OPENVINO_KVCACHE_SPACE` to specify the KV Cache size (e.g, `VLLM_OPENVINO_KVCACHE_SPACE=40` means 40 GB space for KV cache), larger setting will allow vLLM running more requests in parallel. This parameter should be set based on the hardware configuration and memory management pattern of users.
- `VLLM_OPENVINO_CPU_KV_CACHE_PRECISION=u8` to control KV cache precision. By default, FP16 / BF16 is used depending on platform.
To enable better TPOT / TTFT latency, you can use vLLM's chunked prefill feature (`--enable-chunked-prefill`). Based on the experiments, the recommended batch size is `256` (`--max-num-batched-tokens`)
OpenVINO best known configuration for CPU is:
```console
$ VLLM_OPENVINO_KVCACHE_SPACE=100 VLLM_OPENVINO_CPU_KV_CACHE_PRECISION=u8 VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS=ON \
python3 vllm/benchmarks/benchmark_throughput.py --model meta-llama/Llama-2-7b-chat-hf --dataset vllm/benchmarks/ShareGPT_V3_unfiltered_cleaned_split.json --enable-chunked-prefill --max-num-batched-tokens 256
```
### GPU performance tips
GPU device implements the logic for automatic detection of available GPU memory and, by default, tries to reserve as much memory as possible for the KV cache (taking into account `gpu_memory_utilization` option). However, this behavior can be overridden by explicitly specifying the desired amount of memory for the KV cache using `VLLM_OPENVINO_KVCACHE_SPACE` environment variable (e.g, `VLLM_OPENVINO_KVCACHE_SPACE=8` means 8 GB space for KV cache).
Currently, the best performance using GPU can be achieved with the default vLLM execution parameters for models with quantized weights (8 and 4-bit integer data types are supported) and `preemption-mode=swap`.
OpenVINO best known configuration for GPU is:
```console
$ VLLM_OPENVINO_DEVICE=GPU VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS=ON \
python3 vllm/benchmarks/benchmark_throughput.py --model meta-llama/Llama-2-7b-chat-hf --dataset vllm/benchmarks/ShareGPT_V3_unfiltered_cleaned_split.json
```
(openvino-backend-limitations)=
## Limitations
- LoRA serving is not supported.
- Only LLM models are currently supported. LLaVa and encoder-decoder models are not currently enabled in vLLM OpenVINO integration.
- Tensor and pipeline parallelism are not currently enabled in vLLM integration.

116
docs/source/getting_started/openvino-installation.rst
View File

@ -1,116 +0,0 @@
.. _installation_openvino:
Installation with OpenVINO
==========================
vLLM powered by OpenVINO supports all LLM models from :doc:`vLLM supported models list <../models/supported_models>` and can perform optimal model serving on all x86-64 CPUs with, at least, AVX2 support, as well as on both integrated and discrete Intel® GPUs (`the list of supported GPUs <https://docs.openvino.ai/2024/about-openvino/release-notes-openvino/system-requirements.html#gpu>`_). OpenVINO vLLM backend supports the following advanced vLLM features:
- Prefix caching (``--enable-prefix-caching``)
- Chunked prefill (``--enable-chunked-prefill``)
**Table of contents**:
- :ref:`Requirements <openvino_backend_requirements>`
- :ref:`Quick start using Dockerfile <openvino_backend_quick_start_dockerfile>`
- :ref:`Build from source <install_openvino_backend_from_source>`
- :ref:`Performance tips <openvino_backend_performance_tips>`
- :ref:`Limitations <openvino_backend_limitations>`
.. _openvino_backend_requirements:
Requirements
------------
* OS: Linux
* Instruction set architecture (ISA) requirement: at least AVX2.
.. _openvino_backend_quick_start_dockerfile:
Quick start using Dockerfile
----------------------------
.. code-block:: console
$ docker build -f Dockerfile.openvino -t vllm-openvino-env .
$ docker run -it --rm vllm-openvino-env
.. _install_openvino_backend_from_source:
Install from source
-------------------
- First, install Python. For example, on Ubuntu 22.04, you can run:
.. code-block:: console
$ sudo apt-get update -y
$ sudo apt-get install python3
- Second, install prerequisites vLLM OpenVINO backend installation:
.. code-block:: console
$ pip install --upgrade pip
$ pip install -r requirements-build.txt --extra-index-url https://download.pytorch.org/whl/cpu
- Finally, install vLLM with OpenVINO backend:
.. code-block:: console
$ PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu" VLLM_TARGET_DEVICE=openvino python -m pip install -v .
- [Optional] To use vLLM OpenVINO backend with a GPU device, ensure your system is properly set up. Follow the instructions provided here: `https://docs.openvino.ai/2024/get-started/configurations/configurations-intel-gpu.html <https://docs.openvino.ai/2024/get-started/configurations/configurations-intel-gpu.html>`_.
.. _openvino_backend_performance_tips:
Performance tips
----------------
vLLM OpenVINO backend environment variables
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- ``VLLM_OPENVINO_DEVICE`` to specify which device utilize for the inference. If there are multiple GPUs in the system, additional indexes can be used to choose the proper one (e.g, ``VLLM_OPENVINO_DEVICE=GPU.1``). If the value is not specified, CPU device is used by default.
- ``VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS=ON`` to enable U8 weights compression during model loading stage. By default, compression is turned off. You can also export model with different compression techniques using `optimum-cli` and pass exported folder as `<model_id>`
CPU performance tips
~~~~~~~~~~~~~~~~~~~~
CPU uses the following environment variables to control behavior:
- ``VLLM_OPENVINO_KVCACHE_SPACE`` to specify the KV Cache size (e.g, ``VLLM_OPENVINO_KVCACHE_SPACE=40`` means 40 GB space for KV cache), larger setting will allow vLLM running more requests in parallel. This parameter should be set based on the hardware configuration and memory management pattern of users.
- ``VLLM_OPENVINO_CPU_KV_CACHE_PRECISION=u8`` to control KV cache precision. By default, FP16 / BF16 is used depending on platform.
To enable better TPOT / TTFT latency, you can use vLLM's chunked prefill feature (``--enable-chunked-prefill``). Based on the experiments, the recommended batch size is ``256`` (``--max-num-batched-tokens``)
OpenVINO best known configuration for CPU is:
.. code-block:: console
$ VLLM_OPENVINO_KVCACHE_SPACE=100 VLLM_OPENVINO_CPU_KV_CACHE_PRECISION=u8 VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS=ON \
python3 vllm/benchmarks/benchmark_throughput.py --model meta-llama/Llama-2-7b-chat-hf --dataset vllm/benchmarks/ShareGPT_V3_unfiltered_cleaned_split.json --enable-chunked-prefill --max-num-batched-tokens 256
GPU performance tips
~~~~~~~~~~~~~~~~~~~~
GPU device implements the logic for automatic detection of available GPU memory and, by default, tries to reserve as much memory as possible for the KV cache (taking into account ``gpu_memory_utilization`` option). However, this behavior can be overridden by explicitly specifying the desired amount of memory for the KV cache using ``VLLM_OPENVINO_KVCACHE_SPACE`` environment variable (e.g, ``VLLM_OPENVINO_KVCACHE_SPACE=8`` means 8 GB space for KV cache).
Currently, the best performance using GPU can be achieved with the default vLLM execution parameters for models with quantized weights (8 and 4-bit integer data types are supported) and `preemption-mode=swap`.
OpenVINO best known configuration for GPU is:
.. code-block:: console
$ VLLM_OPENVINO_DEVICE=GPU VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS=ON \
python3 vllm/benchmarks/benchmark_throughput.py --model meta-llama/Llama-2-7b-chat-hf --dataset vllm/benchmarks/ShareGPT_V3_unfiltered_cleaned_split.json
.. _openvino_backend_limitations:
Limitations
-----------
- LoRA serving is not supported.
- Only LLM models are currently supported. LLaVa and encoder-decoder models are not currently enabled in vLLM OpenVINO integration.
- Tensor and pipeline parallelism are not currently enabled in vLLM integration.

174
docs/source/getting_started/quickstart.md Normal file
View File

@ -0,0 +1,174 @@
(quickstart)=
# Quickstart
This guide will help you quickly get started with vLLM to:
- [Run offline batched inference](#offline-batched-inference)
- [Run OpenAI-compatible inference](#openai-compatible-server)
## Prerequisites
- OS: Linux
- Python: 3.9 -- 3.12
- GPU: compute capability 7.0 or higher (e.g., V100, T4, RTX20xx, A100, L4, H100, etc.)
## Installation
You can install vLLM using pip. It's recommended to use [conda](https://docs.conda.io/projects/conda/en/latest/user-guide/getting-started.html) to create and manage Python environments.
```console
$ conda create -n myenv python=3.10 -y
$ conda activate myenv
$ pip install vllm
```
Please refer to the {ref}`installation documentation <installation>` for more details on installing vLLM.
(offline-batched-inference)=
## Offline Batched Inference
With vLLM installed, you can start generating texts for list of input prompts (i.e. offline batch inferencing). The example script for this section can be found [here](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference.py).
The first line of this example imports the classes {class}`~vllm.LLM` and {class}`~vllm.SamplingParams`:
- {class}`~vllm.LLM` is the main class for running offline inference with vLLM engine.
- {class}`~vllm.SamplingParams` specifies the parameters for the sampling process.
```python
from vllm import LLM, SamplingParams
```
The next section defines a list of input prompts and sampling parameters for text generation. The [sampling temperature](https://arxiv.org/html/2402.05201v1) is set to `0.8` and the [nucleus sampling probability](https://en.wikipedia.org/wiki/Top-p_sampling) is set to `0.95`. You can find more information about the sampling parameters [here](https://docs.vllm.ai/en/stable/dev/sampling_params.html).
```python
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
```
The {class}`~vllm.LLM` class initializes vLLM's engine and the [OPT-125M model](https://arxiv.org/abs/2205.01068) for offline inference. The list of supported models can be found [here](#supported-models).
```python
llm = LLM(model="facebook/opt-125m")
```
```{note}
By default, vLLM downloads models from [HuggingFace](https://huggingface.co/). If you would like to use models from [ModelScope](https://www.modelscope.cn), set the environment variable `VLLM_USE_MODELSCOPE` before initializing the engine.
```
Now, the fun part! The outputs are generated using `llm.generate`. It adds the input prompts to the vLLM engine's waiting queue and executes the vLLM engine to generate the outputs with high throughput. The outputs are returned as a list of `RequestOutput` objects, which include all of the output tokens.
```python
outputs = llm.generate(prompts, sampling_params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
(openai-compatible-server)=
## OpenAI-Compatible Server
vLLM can be deployed as a server that implements the OpenAI API protocol. This allows vLLM to be used as a drop-in replacement for applications using OpenAI API.
By default, it starts the server at `http://localhost:8000`. You can specify the address with `--host` and `--port` arguments. The server currently hosts one model at a time and implements endpoints such as [list models](https://platform.openai.com/docs/api-reference/models/list), [create chat completion](https://platform.openai.com/docs/api-reference/chat/completions/create), and [create completion](https://platform.openai.com/docs/api-reference/completions/create) endpoints.
Run the following command to start the vLLM server with the [Qwen2.5-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct) model:
```console
$ vllm serve Qwen/Qwen2.5-1.5B-Instruct
```
```{note}
By default, the server uses a predefined chat template stored in the tokenizer. You can learn about overriding it [here](https://github.com/vllm-project/vllm/blob/main/docs/source/serving/openai_compatible_server.md#chat-template).
```
This server can be queried in the same format as OpenAI API. For example, to list the models:
```console
$ curl http://localhost:8000/v1/models
```
You can pass in the argument `--api-key` or environment variable `VLLM_API_KEY` to enable the server to check for API key in the header.
### OpenAI Completions API with vLLM
Once your server is started, you can query the model with input prompts:
```console
$ curl http://localhost:8000/v1/completions \
$ -H "Content-Type: application/json" \
$ -d '{
$ "model": "Qwen/Qwen2.5-1.5B-Instruct",
$ "prompt": "San Francisco is a",
$ "max_tokens": 7,
$ "temperature": 0
$ }'
```
Since this server is compatible with OpenAI API, you can use it as a drop-in replacement for any applications using OpenAI API. For example, another way to query the server is via the `openai` python package:
```python
from openai import OpenAI
# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
completion = client.completions.create(model="Qwen/Qwen2.5-1.5B-Instruct",
prompt="San Francisco is a")
print("Completion result:", completion)
```
A more detailed client example can be found [here](https://github.com/vllm-project/vllm/blob/main/examples/openai_completion_client.py).
### OpenAI Chat Completions API with vLLM
vLLM is designed to also support the OpenAI Chat Completions API. The chat interface is a more dynamic, interactive way to communicate with the model, allowing back-and-forth exchanges that can be stored in the chat history. This is useful for tasks that require context or more detailed explanations.
You can use the [create chat completion](https://platform.openai.com/docs/api-reference/chat/completions/create) endpoint to interact with the model:
```console
$ curl http://localhost:8000/v1/chat/completions \
$ -H "Content-Type: application/json" \
$ -d '{
$ "model": "Qwen/Qwen2.5-1.5B-Instruct",
$ "messages": [
$ {"role": "system", "content": "You are a helpful assistant."},
$ {"role": "user", "content": "Who won the world series in 2020?"}
$ ]
$ }'
```
Alternatively, you can use the `openai` python package:
```python
from openai import OpenAI
# Set OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
chat_response = client.chat.completions.create(
model="Qwen/Qwen2.5-1.5B-Instruct",
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Tell me a joke."},
]
)
print("Chat response:", chat_response)
```

181
docs/source/getting_started/quickstart.rst
View File

@ -1,181 +0,0 @@
.. _quickstart:
==========
Quickstart
==========
This guide will help you quickly get started with vLLM to:
* :ref:`Run offline batched inference <offline_batched_inference>`
* :ref:`Run OpenAI-compatible inference <openai_compatible_server>`
Prerequisites
--------------
- OS: Linux
- Python: 3.9 -- 3.12
- GPU: compute capability 7.0 or higher (e.g., V100, T4, RTX20xx, A100, L4, H100, etc.)
Installation
--------------
You can install vLLM using pip. It's recommended to use `conda <https://docs.conda.io/projects/conda/en/latest/user-guide/getting-started.html>`_ to create and manage Python environments.
.. code-block:: console
$ conda create -n myenv python=3.10 -y
$ conda activate myenv
$ pip install vllm
Please refer to the :ref:`installation documentation <installation>` for more details on installing vLLM.
.. _offline_batched_inference:
Offline Batched Inference
-------------------------
With vLLM installed, you can start generating texts for list of input prompts (i.e. offline batch inferencing). The example script for this section can be found `here <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference.py>`__.
The first line of this example imports the classes :class:`~vllm.LLM` and :class:`~vllm.SamplingParams`:
- :class:`~vllm.LLM` is the main class for running offline inference with vLLM engine.
- :class:`~vllm.SamplingParams` specifies the parameters for the sampling process.
.. code-block:: python
from vllm import LLM, SamplingParams
The next section defines a list of input prompts and sampling parameters for text generation. The `sampling temperature <https://arxiv.org/html/2402.05201v1>`_ is set to ``0.8`` and the `nucleus sampling probability <https://en.wikipedia.org/wiki/Top-p_sampling>`_ is set to ``0.95``. You can find more information about the sampling parameters `here <https://docs.vllm.ai/en/stable/dev/sampling_params.html>`__.
.. code-block:: python
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
The :class:`~vllm.LLM` class initializes vLLM's engine and the `OPT-125M model <https://arxiv.org/abs/2205.01068>`_ for offline inference. The list of supported models can be found :ref:`here <supported_models>`.
.. code-block:: python
llm = LLM(model="facebook/opt-125m")
.. note::
By default, vLLM downloads models from `HuggingFace <https://huggingface.co/>`_. If you would like to use models from `ModelScope <https://www.modelscope.cn>`_, set the environment variable ``VLLM_USE_MODELSCOPE`` before initializing the engine.
Now, the fun part! The outputs are generated using ``llm.generate``. It adds the input prompts to the vLLM engine's waiting queue and executes the vLLM engine to generate the outputs with high throughput. The outputs are returned as a list of ``RequestOutput`` objects, which include all of the output tokens.
.. code-block:: python
outputs = llm.generate(prompts, sampling_params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
.. _openai_compatible_server:
OpenAI-Compatible Server
------------------------
vLLM can be deployed as a server that implements the OpenAI API protocol. This allows vLLM to be used as a drop-in replacement for applications using OpenAI API.
By default, it starts the server at ``http://localhost:8000``. You can specify the address with ``--host`` and ``--port`` arguments. The server currently hosts one model at a time and implements endpoints such as `list models <https://platform.openai.com/docs/api-reference/models/list>`_, `create chat completion <https://platform.openai.com/docs/api-reference/chat/completions/create>`_, and `create completion <https://platform.openai.com/docs/api-reference/completions/create>`_ endpoints.
Run the following command to start the vLLM server with the `Qwen2.5-1.5B-Instruct <https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct>`_ model:
.. code-block:: console
$ vllm serve Qwen/Qwen2.5-1.5B-Instruct
.. note::
By default, the server uses a predefined chat template stored in the tokenizer. You can learn about overriding it `here <https://github.com/vllm-project/vllm/blob/main/docs/source/serving/openai_compatible_server.md#chat-template>`__.
This server can be queried in the same format as OpenAI API. For example, to list the models:
.. code-block:: console
$ curl http://localhost:8000/v1/models
You can pass in the argument ``--api-key`` or environment variable ``VLLM_API_KEY`` to enable the server to check for API key in the header.
OpenAI Completions API with vLLM
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Once your server is started, you can query the model with input prompts:
.. code-block:: console
$ curl http://localhost:8000/v1/completions \
$ -H "Content-Type: application/json" \
$ -d '{
$ "model": "Qwen/Qwen2.5-1.5B-Instruct",
$ "prompt": "San Francisco is a",
$ "max_tokens": 7,
$ "temperature": 0
$ }'
Since this server is compatible with OpenAI API, you can use it as a drop-in replacement for any applications using OpenAI API. For example, another way to query the server is via the ``openai`` python package:
.. code-block:: python
from openai import OpenAI
# Modify OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
completion = client.completions.create(model="Qwen/Qwen2.5-1.5B-Instruct",
prompt="San Francisco is a")
print("Completion result:", completion)
A more detailed client example can be found `here <https://github.com/vllm-project/vllm/blob/main/examples/openai_completion_client.py>`__.
OpenAI Chat Completions API with vLLM
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
vLLM is designed to also support the OpenAI Chat Completions API. The chat interface is a more dynamic, interactive way to communicate with the model, allowing back-and-forth exchanges that can be stored in the chat history. This is useful for tasks that require context or more detailed explanations.
You can use the `create chat completion <https://platform.openai.com/docs/api-reference/chat/completions/create>`_ endpoint to interact with the model:
.. code-block:: console
$ curl http://localhost:8000/v1/chat/completions \
$ -H "Content-Type: application/json" \
$ -d '{
$ "model": "Qwen/Qwen2.5-1.5B-Instruct",
$ "messages": [
$ {"role": "system", "content": "You are a helpful assistant."},
$ {"role": "user", "content": "Who won the world series in 2020?"}
$ ]
$ }'
Alternatively, you can use the ``openai`` python package:
.. code-block:: python
from openai import OpenAI
# Set OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
chat_response = client.chat.completions.create(
model="Qwen/Qwen2.5-1.5B-Instruct",
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Tell me a joke."},
]
)
print("Chat response:", chat_response)

193
docs/source/getting_started/tpu-installation.md Normal file
View File

@ -0,0 +1,193 @@
(installation-tpu)=
# Installation with TPU
Tensor Processing Units (TPUs) are Google's custom-developed application-specific
integrated circuits (ASICs) used to accelerate machine learning workloads. TPUs
are available in different versions each with different hardware specifications.
For more information about TPUs, see [TPU System Architecture](https://cloud.google.com/tpu/docs/system-architecture-tpu-vm).
For more information on the TPU versions supported with vLLM, see:
- [TPU v6e](https://cloud.google.com/tpu/docs/v6e)
- [TPU v5e](https://cloud.google.com/tpu/docs/v5e)
- [TPU v5p](https://cloud.google.com/tpu/docs/v5p)
- [TPU v4](https://cloud.google.com/tpu/docs/v4)
These TPU versions allow you to configure the physical arrangements of the TPU
chips. This can improve throughput and networking performance. For more
information see:
- [TPU v6e topologies](https://cloud.google.com/tpu/docs/v6e#configurations)
- [TPU v5e topologies](https://cloud.google.com/tpu/docs/v5e#tpu-v5e-config)
- [TPU v5p topologies](https://cloud.google.com/tpu/docs/v5p#tpu-v5p-config)
- [TPU v4 topologies](https://cloud.google.com/tpu/docs/v4#tpu-v4-config)
In order for you to use Cloud TPUs you need to have TPU quota granted to your
Google Cloud Platform project. TPU quotas specify how many TPUs you can use in a
GPC project and are specified in terms of TPU version, the number of TPU you
want to use, and quota type. For more information, see [TPU quota](https://cloud.google.com/tpu/docs/quota#tpu_quota).
For TPU pricing information, see [Cloud TPU pricing](https://cloud.google.com/tpu/pricing).
You may need additional persistent storage for your TPU VMs. For more
information, see [Storage options for Cloud TPU data](https://cloud.devsite.corp.google.com/tpu/docs/storage-options).
## Requirements
- Google Cloud TPU VM
- TPU versions: v6e, v5e, v5p, v4
- Python: 3.10 or newer
### Provision Cloud TPUs
You can provision Cloud TPUs using the [Cloud TPU API](https://cloud.google.com/tpu/docs/reference/rest)
or the [queued resources](https://cloud.google.com/tpu/docs/queued-resources)
API. This section shows how to create TPUs using the queued resource API. For
more information about using the Cloud TPU API, see [Create a Cloud TPU using the Create Node API](https://cloud.google.com/tpu/docs/managing-tpus-tpu-vm#create-node-api).
Queued resources enable you to request Cloud TPU resources in a queued manner.
When you request queued resources, the request is added to a queue maintained by
the Cloud TPU service. When the requested resource becomes available, it's
assigned to your Google Cloud project for your immediate exclusive use.
```{note}
In all of the following commands, replace the ALL CAPS parameter names with
appropriate values. See the parameter descriptions table for more information.
```
## Provision a Cloud TPU with the queued resource API
Create a TPU v5e with 4 TPU chips:
```console
gcloud alpha compute tpus queued-resources create QUEUED_RESOURCE_ID \
--node-id TPU_NAME \
--project PROJECT_ID \
--zone ZONE \
--accelerator-type ACCELERATOR_TYPE \
--runtime-version RUNTIME_VERSION \
--service-account SERVICE_ACCOUNT
```
```{eval-rst}
.. list-table:: Parameter descriptions
:header-rows: 1
* - Parameter name
- Description
* - QUEUED_RESOURCE_ID
- The user-assigned ID of the queued resource request.
* - TPU_NAME
- The user-assigned name of the TPU which is created when the queued
resource request is allocated.
* - PROJECT_ID
- Your Google Cloud project
* - ZONE
- The GCP zone where you want to create your Cloud TPU. The value you use
depends on the version of TPUs you are using. For more information, see
`TPU regions and zones <https://cloud.google.com/tpu/docs/regions-zones>`_
* - ACCELERATOR_TYPE
- The TPU version you want to use. Specify the TPU version, for example
`v5litepod-4` specifies a v5e TPU with 4 cores. For more information,
see `TPU versions <https://cloud.devsite.corp.google.com/tpu/docs/system-architecture-tpu-vm#versions>`_.
* - RUNTIME_VERSION
- The TPU VM runtime version to use. For more information see `TPU VM images <https://cloud.google.com/tpu/docs/runtimes>`_.
* - SERVICE_ACCOUNT
- The email address for your service account. You can find it in the IAM
Cloud Console under *Service Accounts*. For example:
`tpu-service-account@<your_project_ID>.iam.gserviceaccount.com`
```
Connect to your TPU using SSH:
```bash
gcloud compute tpus tpu-vm ssh TPU_NAME --zone ZONE
```
Install Miniconda
```bash
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc
```
Create and activate a Conda environment for vLLM:
```bash
conda create -n vllm python=3.10 -y
conda activate vllm
```
Clone the vLLM repository and go to the vLLM directory:
```bash
git clone https://github.com/vllm-project/vllm.git && cd vllm
```
Uninstall the existing `torch` and `torch_xla` packages:
```bash
pip uninstall torch torch-xla -y
```
Install build dependencies:
```bash
pip install -r requirements-tpu.txt
sudo apt-get install libopenblas-base libopenmpi-dev libomp-dev
```
Run the setup script:
```bash
VLLM_TARGET_DEVICE="tpu" python setup.py develop
```
## Provision Cloud TPUs with GKE
For more information about using TPUs with GKE, see
<https://cloud.google.com/kubernetes-engine/docs/how-to/tpus>
<https://cloud.google.com/kubernetes-engine/docs/concepts/tpus>
<https://cloud.google.com/kubernetes-engine/docs/concepts/plan-tpus>
(build-docker-tpu)=
## Build a docker image with {code}`Dockerfile.tpu`
You can use [Dockerfile.tpu](https://github.com/vllm-project/vllm/blob/main/Dockerfile.tpu)
to build a Docker image with TPU support.
```console
$ docker build -f Dockerfile.tpu -t vllm-tpu .
```
Run the Docker image with the following command:
```console
$ # Make sure to add `--privileged --net host --shm-size=16G`.
$ docker run --privileged --net host --shm-size=16G -it vllm-tpu
```
```{note}
Since TPU relies on XLA which requires static shapes, vLLM bucketizes the
possible input shapes and compiles an XLA graph for each shape. The
compilation time may take 20~30 minutes in the first run. However, the
compilation time reduces to ~5 minutes afterwards because the XLA graphs are
cached in the disk (in {code}`VLLM_XLA_CACHE_PATH` or {code}`~/.cache/vllm/xla_cache` by default).
```
````{tip}
If you encounter the following error:
```console
from torch._C import * # noqa: F403
ImportError: libopenblas.so.0: cannot open shared object file: No such
file or directory
```
Install OpenBLAS with the following command:
```console
$ sudo apt-get install libopenblas-base libopenmpi-dev libomp-dev
```
````

200
docs/source/getting_started/tpu-installation.rst
View File

@ -1,200 +0,0 @@
.. _installation_tpu:
#####################
Installation with TPU
#####################
Tensor Processing Units (TPUs) are Google's custom-developed application-specific
integrated circuits (ASICs) used to accelerate machine learning workloads. TPUs
are available in different versions each with different hardware specifications.
For more information about TPUs, see `TPU System Architecture <https://cloud.google.com/tpu/docs/system-architecture-tpu-vm>`_.
For more information on the TPU versions supported with vLLM, see:
* `TPU v6e <https://cloud.google.com/tpu/docs/v6e>`_
* `TPU v5e <https://cloud.google.com/tpu/docs/v5e>`_
* `TPU v5p <https://cloud.google.com/tpu/docs/v5p>`_
* `TPU v4 <https://cloud.google.com/tpu/docs/v4>`_
These TPU versions allow you to configure the physical arrangements of the TPU
chips. This can improve throughput and networking performance. For more
information see:
* `TPU v6e topologies <https://cloud.google.com/tpu/docs/v6e#configurations>`_
* `TPU v5e topologies <https://cloud.google.com/tpu/docs/v5e#tpu-v5e-config>`_
* `TPU v5p topologies <https://cloud.google.com/tpu/docs/v5p#tpu-v5p-config>`_
* `TPU v4 topologies <https://cloud.google.com/tpu/docs/v4#tpu-v4-config>`_
In order for you to use Cloud TPUs you need to have TPU quota granted to your
Google Cloud Platform project. TPU quotas specify how many TPUs you can use in a
GPC project and are specified in terms of TPU version, the number of TPU you
want to use, and quota type. For more information, see `TPU quota <https://cloud.google.com/tpu/docs/quota#tpu_quota>`_.
For TPU pricing information, see `Cloud TPU pricing <https://cloud.google.com/tpu/pricing>`_.
You may need additional persistent storage for your TPU VMs. For more
information, see `Storage options for Cloud TPU data <https://cloud.devsite.corp.google.com/tpu/docs/storage-options>`_.
Requirements
------------
* Google Cloud TPU VM
* TPU versions: v6e, v5e, v5p, v4
* Python: 3.10 or newer
Provision Cloud TPUs
====================
You can provision Cloud TPUs using the `Cloud TPU API <https://cloud.google.com/tpu/docs/reference/rest>`_
or the `queued resources <https://cloud.google.com/tpu/docs/queued-resources>`_
API. This section shows how to create TPUs using the queued resource API. For
more information about using the Cloud TPU API, see `Create a Cloud TPU using the Create Node API <https://cloud.google.com/tpu/docs/managing-tpus-tpu-vm#create-node-api>`_.
Queued resources enable you to request Cloud TPU resources in a queued manner.
When you request queued resources, the request is added to a queue maintained by
the Cloud TPU service. When the requested resource becomes available, it's
assigned to your Google Cloud project for your immediate exclusive use.
.. note::
In all of the following commands, replace the ALL CAPS parameter names with
appropriate values. See the parameter descriptions table for more information.
Provision a Cloud TPU with the queued resource API
--------------------------------------------------
Create a TPU v5e with 4 TPU chips:
.. code-block:: console
gcloud alpha compute tpus queued-resources create QUEUED_RESOURCE_ID \
--node-id TPU_NAME \
--project PROJECT_ID \
--zone ZONE \
--accelerator-type ACCELERATOR_TYPE \
--runtime-version RUNTIME_VERSION \
--service-account SERVICE_ACCOUNT
.. list-table:: Parameter descriptions
:header-rows: 1
* - Parameter name
- Description
* - QUEUED_RESOURCE_ID
- The user-assigned ID of the queued resource request.
* - TPU_NAME
- The user-assigned name of the TPU which is created when the queued
resource request is allocated.
* - PROJECT_ID
- Your Google Cloud project
* - ZONE
- The GCP zone where you want to create your Cloud TPU. The value you use
depends on the version of TPUs you are using. For more information, see
`TPU regions and zones <https://cloud.google.com/tpu/docs/regions-zones>`_
* - ACCELERATOR_TYPE
- The TPU version you want to use. Specify the TPU version, for example
`v5litepod-4` specifies a v5e TPU with 4 cores. For more information,
see `TPU versions <https://cloud.devsite.corp.google.com/tpu/docs/system-architecture-tpu-vm#versions>`_.
* - RUNTIME_VERSION
- The TPU VM runtime version to use. For more information see `TPU VM images <https://cloud.google.com/tpu/docs/runtimes>`_.
* - SERVICE_ACCOUNT
- The email address for your service account. You can find it in the IAM
Cloud Console under *Service Accounts*. For example:
`tpu-service-account@<your_project_ID>.iam.gserviceaccount.com`
Connect to your TPU using SSH:
.. code-block:: bash
gcloud compute tpus tpu-vm ssh TPU_NAME --zone ZONE
Install Miniconda
.. code-block:: bash
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc
Create and activate a Conda environment for vLLM:
.. code-block:: bash
conda create -n vllm python=3.10 -y
conda activate vllm
Clone the vLLM repository and go to the vLLM directory:
.. code-block:: bash
git clone https://github.com/vllm-project/vllm.git && cd vllm
Uninstall the existing `torch` and `torch_xla` packages:
.. code-block:: bash
pip uninstall torch torch-xla -y
Install build dependencies:
.. code-block:: bash
pip install -r requirements-tpu.txt
sudo apt-get install libopenblas-base libopenmpi-dev libomp-dev
Run the setup script:
.. code-block:: bash
VLLM_TARGET_DEVICE="tpu" python setup.py develop
Provision Cloud TPUs with GKE
-----------------------------
For more information about using TPUs with GKE, see
https://cloud.google.com/kubernetes-engine/docs/how-to/tpus
https://cloud.google.com/kubernetes-engine/docs/concepts/tpus
https://cloud.google.com/kubernetes-engine/docs/concepts/plan-tpus
.. _build_docker_tpu:
Build a docker image with :code:`Dockerfile.tpu`
------------------------------------------------
You can use `Dockerfile.tpu <https://github.com/vllm-project/vllm/blob/main/Dockerfile.tpu>`_
to build a Docker image with TPU support.
.. code-block:: console
$ docker build -f Dockerfile.tpu -t vllm-tpu .
Run the Docker image with the following command:
.. code-block:: console
$ # Make sure to add `--privileged --net host --shm-size=16G`.
$ docker run --privileged --net host --shm-size=16G -it vllm-tpu
.. note::
Since TPU relies on XLA which requires static shapes, vLLM bucketizes the
possible input shapes and compiles an XLA graph for each shape. The
compilation time may take 20~30 minutes in the first run. However, the
compilation time reduces to ~5 minutes afterwards because the XLA graphs are
cached in the disk (in :code:`VLLM_XLA_CACHE_PATH` or :code:`~/.cache/vllm/xla_cache` by default).
.. tip::
If you encounter the following error:
.. code-block:: console
from torch._C import * # noqa: F403
ImportError: libopenblas.so.0: cannot open shared object file: No such
file or directory
Install OpenBLAS with the following command:
.. code-block:: console
$ sudo apt-get install libopenblas-base libopenmpi-dev libomp-dev

74
docs/source/getting_started/xpu-installation.md Normal file
View File

@ -0,0 +1,74 @@
(installation-xpu)=
# Installation with XPU
vLLM initially supports basic model inferencing and serving on Intel GPU platform.
Table of contents:
1. [Requirements](#xpu-backend-requirements)
2. [Quick start using Dockerfile](#xpu-backend-quick-start-dockerfile)
3. [Build from source](#build-xpu-backend-from-source)
(xpu-backend-requirements)=
## Requirements
- OS: Linux
- Supported Hardware: Intel Data Center GPU, Intel ARC GPU
- OneAPI requirements: oneAPI 2024.2
(xpu-backend-quick-start-dockerfile)=
## Quick start using Dockerfile
```console
$ docker build -f Dockerfile.xpu -t vllm-xpu-env --shm-size=4g .
$ docker run -it \
--rm \
--network=host \
--device /dev/dri \
-v /dev/dri/by-path:/dev/dri/by-path \
vllm-xpu-env
```
(build-xpu-backend-from-source)=
## Build from source
- First, install required driver and intel OneAPI 2024.2 or later.
- Second, install Python packages for vLLM XPU backend building:
```console
$ source /opt/intel/oneapi/setvars.sh
$ pip install --upgrade pip
$ pip install -v -r requirements-xpu.txt
```
- Finally, build and install vLLM XPU backend:
```console
$ VLLM_TARGET_DEVICE=xpu python setup.py install
```
```{note}
- FP16 is the default data type in the current XPU backend. The BF16 data
type will be supported in the future.
```
## Distributed inference and serving
XPU platform supports tensor-parallel inference/serving and also supports pipeline parallel as a beta feature for online serving. We requires Ray as the distributed runtime backend. For example, a reference execution likes following:
```console
$ python -m vllm.entrypoints.openai.api_server \
$ --model=facebook/opt-13b \
$ --dtype=bfloat16 \
$ --device=xpu \
$ --max_model_len=1024 \
$ --distributed-executor-backend=ray \
$ --pipeline-parallel-size=2 \
$ -tp=8
```
By default, a ray instance will be launched automatically if no existing one is detected in system, with `num-gpus` equals to `parallel_config.world_size`. We recommend properly starting a ray cluster before execution, referring helper [script](https://github.com/vllm-project/vllm/tree/main/examples/run_cluster.sh).

80
docs/source/getting_started/xpu-installation.rst
View File

@ -1,80 +0,0 @@
.. _installation_xpu:
Installation with XPU
========================
vLLM initially supports basic model inferencing and serving on Intel GPU platform.
Table of contents:
#. :ref:`Requirements <xpu_backend_requirements>`
#. :ref:`Quick start using Dockerfile <xpu_backend_quick_start_dockerfile>`
#. :ref:`Build from source <build_xpu_backend_from_source>`
.. _xpu_backend_requirements:
Requirements
------------
* OS: Linux
* Supported Hardware: Intel Data Center GPU, Intel ARC GPU
* OneAPI requirements: oneAPI 2024.2
.. _xpu_backend_quick_start_dockerfile:
Quick start using Dockerfile
----------------------------
.. code-block:: console
$ docker build -f Dockerfile.xpu -t vllm-xpu-env --shm-size=4g .
$ docker run -it \
--rm \
--network=host \
--device /dev/dri \
-v /dev/dri/by-path:/dev/dri/by-path \
vllm-xpu-env
.. _build_xpu_backend_from_source:
Build from source
-----------------
- First, install required driver and intel OneAPI 2024.2 or later.
- Second, install Python packages for vLLM XPU backend building:
.. code-block:: console
$ source /opt/intel/oneapi/setvars.sh
$ pip install --upgrade pip
$ pip install -v -r requirements-xpu.txt
- Finally, build and install vLLM XPU backend:
.. code-block:: console
$ VLLM_TARGET_DEVICE=xpu python setup.py install
.. note::
- FP16 is the default data type in the current XPU backend. The BF16 data
type will be supported in the future.
Distributed inference and serving
---------------------------------
XPU platform supports tensor-parallel inference/serving and also supports pipeline parallel as a beta feature for online serving. We requires Ray as the distributed runtime backend. For example, a reference execution likes following:
.. code-block:: console
$ python -m vllm.entrypoints.openai.api_server \
$ --model=facebook/opt-13b \
$ --dtype=bfloat16 \
$ --device=xpu \
$ --max_model_len=1024 \
$ --distributed-executor-backend=ray \
$ --pipeline-parallel-size=2 \
$ -tp=8
By default, a ray instance will be launched automatically if no existing one is detected in system, with ``num-gpus`` equals to ``parallel_config.world_size``. We recommend properly starting a ray cluster before execution, referring helper `script <https://github.com/vllm-project/vllm/tree/main/examples/run_cluster.sh>`_.

200
docs/source/index.md Normal file
View File

@ -0,0 +1,200 @@
# Welcome to vLLM!
```{figure} ./assets/logos/vllm-logo-text-light.png
:align: center
:alt: vLLM
:class: no-scaled-link
:width: 60%
```
```{raw} html
<p style="text-align:center">
<strong>Easy, fast, and cheap LLM serving for everyone
</strong>
</p>
<p style="text-align:center">
<script async defer src="https://buttons.github.io/buttons.js"></script>
<a class="github-button" href="https://github.com/vllm-project/vllm" data-show-count="true" data-size="large" aria-label="Star">Star</a>
<a class="github-button" href="https://github.com/vllm-project/vllm/subscription" data-icon="octicon-eye" data-size="large" aria-label="Watch">Watch</a>
<a class="github-button" href="https://github.com/vllm-project/vllm/fork" data-icon="octicon-repo-forked" data-size="large" aria-label="Fork">Fork</a>
</p>
```
vLLM is a fast and easy-to-use library for LLM inference and serving.
vLLM is fast with:
- State-of-the-art serving throughput
- Efficient management of attention key and value memory with **PagedAttention**
- Continuous batching of incoming requests
- Fast model execution with CUDA/HIP graph
- Quantization: [GPTQ](https://arxiv.org/abs/2210.17323), [AWQ](https://arxiv.org/abs/2306.00978), INT4, INT8, and FP8
- Optimized CUDA kernels, including integration with FlashAttention and FlashInfer.
- Speculative decoding
- Chunked prefill
vLLM is flexible and easy to use with:
- Seamless integration with popular HuggingFace models
- High-throughput serving with various decoding algorithms, including *parallel sampling*, *beam search*, and more
- Tensor parallelism and pipeline parallelism support for distributed inference
- Streaming outputs
- OpenAI-compatible API server
- Support NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs, Gaudi® accelerators and GPUs, PowerPC CPUs, TPU, and AWS Trainium and Inferentia Accelerators.
- Prefix caching support
- Multi-lora support
For more information, check out the following:
- [vLLM announcing blog post](https://vllm.ai) (intro to PagedAttention)
- [vLLM paper](https://arxiv.org/abs/2309.06180) (SOSP 2023)
- [How continuous batching enables 23x throughput in LLM inference while reducing p50 latency](https://www.anyscale.com/blog/continuous-batching-llm-inference) by Cade Daniel et al.
- {ref}`vLLM Meetups <meetups>`.
## Documentation
```{toctree}
:caption: Getting Started
:maxdepth: 1
getting_started/installation
getting_started/amd-installation
getting_started/openvino-installation
getting_started/cpu-installation
getting_started/gaudi-installation
getting_started/arm-installation
getting_started/neuron-installation
getting_started/tpu-installation
getting_started/xpu-installation
getting_started/quickstart
getting_started/debugging
getting_started/examples/examples_index
```
```{toctree}
:caption: Serving
:maxdepth: 1
serving/openai_compatible_server
serving/deploying_with_docker
serving/deploying_with_k8s
serving/deploying_with_helm
serving/deploying_with_nginx
serving/distributed_serving
serving/metrics
serving/integrations
serving/tensorizer
serving/runai_model_streamer
```
```{toctree}
:caption: Models
:maxdepth: 1
models/supported_models
models/generative_models
models/pooling_models
models/adding_model
models/enabling_multimodal_inputs
```
```{toctree}
:caption: Usage
:maxdepth: 1
usage/lora
usage/multimodal_inputs
usage/tool_calling
usage/structured_outputs
usage/spec_decode
usage/compatibility_matrix
usage/performance
usage/faq
usage/engine_args
usage/env_vars
usage/usage_stats
usage/disagg_prefill
```
```{toctree}
:caption: Quantization
:maxdepth: 1
quantization/supported_hardware
quantization/auto_awq
quantization/bnb
quantization/gguf
quantization/int8
quantization/fp8
quantization/fp8_e5m2_kvcache
quantization/fp8_e4m3_kvcache
```
```{toctree}
:caption: Automatic Prefix Caching
:maxdepth: 1
automatic_prefix_caching/apc
automatic_prefix_caching/details
```
```{toctree}
:caption: Performance
:maxdepth: 1
performance/benchmarks
```
% Community: User community resources
```{toctree}
:caption: Community
:maxdepth: 1
community/meetups
community/sponsors
```
% API Documentation: API reference aimed at vllm library usage
```{toctree}
:caption: API Documentation
:maxdepth: 2
dev/sampling_params
dev/pooling_params
dev/offline_inference/offline_index
dev/engine/engine_index
```
% Design: docs about vLLM internals
```{toctree}
:caption: Design
:maxdepth: 2
design/arch_overview
design/huggingface_integration
design/plugin_system
design/input_processing/model_inputs_index
design/kernel/paged_attention
design/multimodal/multimodal_index
design/multiprocessing
```
% For Developers: contributing to the vLLM project
```{toctree}
:caption: For Developers
:maxdepth: 2
contributing/overview
contributing/profiling/profiling_index
contributing/dockerfile/dockerfile
```
# Indices and tables
- {ref}`genindex`
- {ref}`modindex`

194
docs/source/index.rst
View File

@ -1,194 +0,0 @@
Welcome to vLLM!
================
.. figure:: ./assets/logos/vllm-logo-text-light.png
:width: 60%
:align: center
:alt: vLLM
:class: no-scaled-link
.. raw:: html
<p style="text-align:center">
<strong>Easy, fast, and cheap LLM serving for everyone
</strong>
</p>
<p style="text-align:center">
<script async defer src="https://buttons.github.io/buttons.js"></script>
<a class="github-button" href="https://github.com/vllm-project/vllm" data-show-count="true" data-size="large" aria-label="Star">Star</a>
<a class="github-button" href="https://github.com/vllm-project/vllm/subscription" data-icon="octicon-eye" data-size="large" aria-label="Watch">Watch</a>
<a class="github-button" href="https://github.com/vllm-project/vllm/fork" data-icon="octicon-repo-forked" data-size="large" aria-label="Fork">Fork</a>
</p>
vLLM is a fast and easy-to-use library for LLM inference and serving.
vLLM is fast with:
* State-of-the-art serving throughput
* Efficient management of attention key and value memory with **PagedAttention**
* Continuous batching of incoming requests
* Fast model execution with CUDA/HIP graph
* Quantization: `GPTQ <https://arxiv.org/abs/2210.17323>`_, `AWQ <https://arxiv.org/abs/2306.00978>`_, INT4, INT8, and FP8
* Optimized CUDA kernels, including integration with FlashAttention and FlashInfer.
* Speculative decoding
* Chunked prefill
vLLM is flexible and easy to use with:
* Seamless integration with popular HuggingFace models
* High-throughput serving with various decoding algorithms, including *parallel sampling*, *beam search*, and more
* Tensor parallelism and pipeline parallelism support for distributed inference
* Streaming outputs
* OpenAI-compatible API server
* Support NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs, Gaudi® accelerators and GPUs, PowerPC CPUs, TPU, and AWS Trainium and Inferentia Accelerators.
* Prefix caching support
* Multi-lora support
For more information, check out the following:
* `vLLM announcing blog post <https://vllm.ai>`_ (intro to PagedAttention)
* `vLLM paper <https://arxiv.org/abs/2309.06180>`_ (SOSP 2023)
* `How continuous batching enables 23x throughput in LLM inference while reducing p50 latency <https://www.anyscale.com/blog/continuous-batching-llm-inference>`_ by Cade Daniel et al.
* :ref:`vLLM Meetups <meetups>`.
Documentation
-------------
.. toctree::
:maxdepth: 1
:caption: Getting Started
getting_started/installation
getting_started/amd-installation
getting_started/openvino-installation
getting_started/cpu-installation
getting_started/gaudi-installation
getting_started/arm-installation
getting_started/neuron-installation
getting_started/tpu-installation
getting_started/xpu-installation
getting_started/quickstart
getting_started/debugging
getting_started/examples/examples_index
.. toctree::
:maxdepth: 1
:caption: Serving
serving/openai_compatible_server
serving/deploying_with_docker
serving/deploying_with_k8s
serving/deploying_with_helm
serving/deploying_with_nginx
serving/distributed_serving
serving/metrics
serving/integrations
serving/tensorizer
serving/runai_model_streamer
.. toctree::
:maxdepth: 1
:caption: Models
models/supported_models
models/generative_models
models/pooling_models
models/adding_model
models/enabling_multimodal_inputs
.. toctree::
:maxdepth: 1
:caption: Usage
usage/lora
usage/multimodal_inputs
usage/tool_calling
usage/structured_outputs
usage/spec_decode
usage/compatibility_matrix
usage/performance
usage/faq
usage/engine_args
usage/env_vars
usage/usage_stats
usage/disagg_prefill
.. toctree::
:maxdepth: 1
:caption: Quantization
quantization/supported_hardware
quantization/auto_awq
quantization/bnb
quantization/gguf
quantization/int8
quantization/fp8
quantization/fp8_e5m2_kvcache
quantization/fp8_e4m3_kvcache
.. toctree::
:maxdepth: 1
:caption: Automatic Prefix Caching
automatic_prefix_caching/apc
automatic_prefix_caching/details
.. toctree::
:maxdepth: 1
:caption: Performance
performance/benchmarks
.. Community: User community resources
.. toctree::
:maxdepth: 1
:caption: Community
community/meetups
community/sponsors
.. API Documentation: API reference aimed at vllm library usage
.. toctree::
:maxdepth: 2
:caption: API Documentation
dev/sampling_params
dev/pooling_params
dev/offline_inference/offline_index
dev/engine/engine_index
.. Design: docs about vLLM internals
.. toctree::
:maxdepth: 2
:caption: Design
design/arch_overview
design/huggingface_integration
design/plugin_system
design/input_processing/model_inputs_index
design/kernel/paged_attention
design/multimodal/multimodal_index
design/multiprocessing
.. For Developers: contributing to the vLLM project
.. toctree::
:maxdepth: 2
:caption: For Developers
contributing/overview
contributing/profiling/profiling_index
contributing/dockerfile/dockerfile
Indices and tables
==================
* :ref:`genindex`
* :ref:`modindex`

155
docs/source/models/adding_model.md Normal file
View File

@ -0,0 +1,155 @@
(adding-a-new-model)=
# Adding a New Model
This document provides a high-level guide on integrating a [HuggingFace Transformers](https://github.com/huggingface/transformers) model into vLLM.
```{note}
The complexity of adding a new model depends heavily on the model's architecture.
The process is considerably straightforward if the model shares a similar architecture with an existing model in vLLM.
However, for models that include new operators (e.g., a new attention mechanism), the process can be a bit more complex.
```
```{note}
By default, vLLM models do not support multi-modal inputs. To enable multi-modal support,
please follow [this guide](#enabling-multimodal-inputs) after implementing the model here.
```
```{tip}
If you are encountering issues while integrating your model into vLLM, feel free to open an issue on our [GitHub](https://github.com/vllm-project/vllm/issues) repository.
We will be happy to help you out!
```
## 0. Fork the vLLM repository
Start by forking our [GitHub] repository and then [build it from source](#build-from-source).
This gives you the ability to modify the codebase and test your model.
```{tip}
If you don't want to fork the repository and modify vLLM's codebase, please refer to the "Out-of-Tree Model Integration" section below.
```
## 1. Bring your model code
Clone the PyTorch model code from the HuggingFace Transformers repository and put it into the [vllm/model_executor/models](https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models) directory.
For instance, vLLM's [OPT model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/opt.py) was adapted from the HuggingFace's [modeling_opt.py](https://github.com/huggingface/transformers/blob/main/src/transformers/models/opt/modeling_opt.py) file.
```{warning}
When copying the model code, make sure to review and adhere to the code's copyright and licensing terms.
```
## 2. Make your code compatible with vLLM
To ensure compatibility with vLLM, your model must meet the following requirements:
### Initialization Code
All vLLM modules within the model must include a `prefix` argument in their constructor. This `prefix` is typically the full name of the module in the model's state dictionary and is crucial for:
- Runtime support: vLLM's attention operators are registered in a model's state by their full names. Each attention operator must have a unique prefix as its layer name to avoid conflicts.
- Non-uniform quantization support: A quantized checkpoint can selectively quantize certain layers while keeping others in full precision. By providing the `prefix` during initialization, vLLM can match the current layer's `prefix` with the quantization configuration to determine if the layer should be initialized in quantized mode.
The initialization code should look like this:
```python
from torch import nn
from vllm.config import VllmConfig
from vllm.attention import Attention
class MyAttention(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str):
super().__init__()
self.attn = Attention(prefix=f"{prefix}.attn")
class MyDecoderLayer(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str):
super().__init__()
self.self_attn = MyAttention(prefix=f"{prefix}.self_attn")
class MyModel(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str):
super().__init__()
self.layers = nn.ModuleList(
[MyDecoderLayer(vllm_config, prefix=f"{prefix}.layers.{i}") for i in range(vllm_config.model_config.hf_config.num_hidden_layers)]
)
class MyModelForCausalLM(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
self.model = MyModel(vllm_config, prefix=f"{prefix}.model")
```
### Computation Code
Rewrite the {meth}`~torch.nn.Module.forward` method of your model to remove any unnecessary code, such as training-specific code. Modify the input parameters to treat `input_ids` and `positions` as flattened tensors with a single batch size dimension, without a max-sequence length dimension.
```python
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
) -> torch.Tensor:
...
```
```{note}
Currently, vLLM supports the basic multi-head attention mechanism and its variant with rotary positional embeddings.
If your model employs a different attention mechanism, you will need to implement a new attention layer in vLLM.
```
For reference, check out the [LLAMA model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llama.py). vLLM already supports a large number of models. It is recommended to find a model similar to yours and adapt it to your model's architecture. Check out the [vLLM models](https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models) directory for more examples.
## 3. (Optional) Implement tensor parallelism and quantization support
If your model is too large to fit into a single GPU, you can use tensor parallelism to manage it.
To do this, substitute your model's linear and embedding layers with their tensor-parallel versions.
For the embedding layer, you can simply replace {class}`torch.nn.Embedding` with {code}`VocabParallelEmbedding`. For the output LM head, you can use {code}`ParallelLMHead`.
When it comes to the linear layers, we provide the following options to parallelize them:
- {code}`ReplicatedLinear`: Replicates the inputs and weights across multiple GPUs. No memory saving.
- {code}`RowParallelLinear`: The input tensor is partitioned along the hidden dimension. The weight matrix is partitioned along the rows (input dimension). An *all-reduce* operation is performed after the matrix multiplication to reduce the results. Typically used for the second FFN layer and the output linear transformation of the attention layer.
- {code}`ColumnParallelLinear`: The input tensor is replicated. The weight matrix is partitioned along the columns (output dimension). The result is partitioned along the column dimension. Typically used for the first FFN layer and the separated QKV transformation of the attention layer in the original Transformer.
- {code}`MergedColumnParallelLinear`: Column-parallel linear that merges multiple {code}`ColumnParallelLinear` operators. Typically used for the first FFN layer with weighted activation functions (e.g., SiLU). This class handles the sharded weight loading logic of multiple weight matrices.
- {code}`QKVParallelLinear`: Parallel linear layer for the query, key, and value projections of the multi-head and grouped-query attention mechanisms. When number of key/value heads are less than the world size, this class replicates the key/value heads properly. This class handles the weight loading and replication of the weight matrices.
Note that all the linear layers above take {code}`linear_method` as an input. vLLM will set this parameter according to different quantization schemes to support weight quantization.
## 4. Implement the weight loading logic
You now need to implement the {code}`load_weights` method in your {code}`*ForCausalLM` class.
This method should load the weights from the HuggingFace's checkpoint file and assign them to the corresponding layers in your model. Specifically, for {code}`MergedColumnParallelLinear` and {code}`QKVParallelLinear` layers, if the original model has separated weight matrices, you need to load the different parts separately.
## 5. Register your model
Finally, register your {code}`*ForCausalLM` class to the {code}`_VLLM_MODELS` in [vllm/model_executor/models/registry.py](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/registry.py).
## 6. Out-of-Tree Model Integration
You can integrate a model without modifying the vLLM codebase. Steps 2, 3, and 4 are still required, but you can skip steps 1 and 5. Instead, write a plugin to register your model. For general introduction of the plugin system, see [plugin-system](#plugin-system).
To register the model, use the following code:
```python
from vllm import ModelRegistry
from your_code import YourModelForCausalLM
ModelRegistry.register_model("YourModelForCausalLM", YourModelForCausalLM)
```
If your model imports modules that initialize CUDA, consider lazy-importing it to avoid errors like {code}`RuntimeError: Cannot re-initialize CUDA in forked subprocess`:
```python
from vllm import ModelRegistry
ModelRegistry.register_model("YourModelForCausalLM", "your_code:YourModelForCausalLM")
```
```{important}
If your model is a multimodal model, ensure the model class implements the {class}`~vllm.model_executor.models.interfaces.SupportsMultiModal` interface.
Read more about that [here](#enabling-multimodal-inputs).
```
```{note}
Although you can directly put these code snippets in your script using `vllm.LLM`, the recommended way is to place these snippets in a vLLM plugin. This ensures compatibility with various vLLM features like distributed inference and the API server.
```

159
docs/source/models/adding_model.rst
View File

@ -1,159 +0,0 @@
.. _adding_a_new_model:
Adding a New Model
==================
This document provides a high-level guide on integrating a `HuggingFace Transformers <https://github.com/huggingface/transformers>`_ model into vLLM.
.. note::
The complexity of adding a new model depends heavily on the model's architecture.
The process is considerably straightforward if the model shares a similar architecture with an existing model in vLLM.
However, for models that include new operators (e.g., a new attention mechanism), the process can be a bit more complex.
.. note::
By default, vLLM models do not support multi-modal inputs. To enable multi-modal support,
please follow :ref:`this guide <enabling_multimodal_inputs>` after implementing the model here.
.. tip::
If you are encountering issues while integrating your model into vLLM, feel free to open an issue on our `GitHub <https://github.com/vllm-project/vllm/issues>`_ repository.
We will be happy to help you out!
0. Fork the vLLM repository
--------------------------------
Start by forking our `GitHub`_ repository and then :ref:`build it from source <build_from_source>`.
This gives you the ability to modify the codebase and test your model.
.. tip::
If you don't want to fork the repository and modify vLLM's codebase, please refer to the "Out-of-Tree Model Integration" section below.
1. Bring your model code
------------------------
Clone the PyTorch model code from the HuggingFace Transformers repository and put it into the `vllm/model_executor/models <https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models>`_ directory.
For instance, vLLM's `OPT model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/opt.py>`_ was adapted from the HuggingFace's `modeling_opt.py <https://github.com/huggingface/transformers/blob/main/src/transformers/models/opt/modeling_opt.py>`_ file.
.. warning::
When copying the model code, make sure to review and adhere to the code's copyright and licensing terms.
2. Make your code compatible with vLLM
--------------------------------------
To ensure compatibility with vLLM, your model must meet the following requirements:
Initialization Code
^^^^^^^^^^^^^^^^^^^
All vLLM modules within the model must include a ``prefix`` argument in their constructor. This ``prefix`` is typically the full name of the module in the model's state dictionary and is crucial for:
* Runtime support: vLLM's attention operators are registered in a model's state by their full names. Each attention operator must have a unique prefix as its layer name to avoid conflicts.
* Non-uniform quantization support: A quantized checkpoint can selectively quantize certain layers while keeping others in full precision. By providing the ``prefix`` during initialization, vLLM can match the current layer's ``prefix`` with the quantization configuration to determine if the layer should be initialized in quantized mode.
The initialization code should look like this:
.. code-block:: python
from torch import nn
from vllm.config import VllmConfig
from vllm.attention import Attention
class MyAttention(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str):
super().__init__()
self.attn = Attention(prefix=f"{prefix}.attn")
class MyDecoderLayer(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str):
super().__init__()
self.self_attn = MyAttention(prefix=f"{prefix}.self_attn")
class MyModel(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str):
super().__init__()
self.layers = nn.ModuleList(
[MyDecoderLayer(vllm_config, prefix=f"{prefix}.layers.{i}") for i in range(vllm_config.model_config.hf_config.num_hidden_layers)]
)
class MyModelForCausalLM(nn.Module):
def __init__(self, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
self.model = MyModel(vllm_config, prefix=f"{prefix}.model")
Computation Code
^^^^^^^^^^^^^^^^
Rewrite the :meth:`~torch.nn.Module.forward` method of your model to remove any unnecessary code, such as training-specific code. Modify the input parameters to treat ``input_ids`` and ``positions`` as flattened tensors with a single batch size dimension, without a max-sequence length dimension.
.. code-block:: python
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
) -> torch.Tensor:
...
.. note::
Currently, vLLM supports the basic multi-head attention mechanism and its variant with rotary positional embeddings.
If your model employs a different attention mechanism, you will need to implement a new attention layer in vLLM.
For reference, check out the `LLAMA model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llama.py>`__. vLLM already supports a large number of models. It is recommended to find a model similar to yours and adapt it to your model's architecture. Check out the `vLLM models <https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/models>`__ directory for more examples.
3. (Optional) Implement tensor parallelism and quantization support
-------------------------------------------------------------------
If your model is too large to fit into a single GPU, you can use tensor parallelism to manage it.
To do this, substitute your model's linear and embedding layers with their tensor-parallel versions.
For the embedding layer, you can simply replace :class:`torch.nn.Embedding` with :code:`VocabParallelEmbedding`. For the output LM head, you can use :code:`ParallelLMHead`.
When it comes to the linear layers, we provide the following options to parallelize them:
* :code:`ReplicatedLinear`: Replicates the inputs and weights across multiple GPUs. No memory saving.
* :code:`RowParallelLinear`: The input tensor is partitioned along the hidden dimension. The weight matrix is partitioned along the rows (input dimension). An *all-reduce* operation is performed after the matrix multiplication to reduce the results. Typically used for the second FFN layer and the output linear transformation of the attention layer.
* :code:`ColumnParallelLinear`: The input tensor is replicated. The weight matrix is partitioned along the columns (output dimension). The result is partitioned along the column dimension. Typically used for the first FFN layer and the separated QKV transformation of the attention layer in the original Transformer.
* :code:`MergedColumnParallelLinear`: Column-parallel linear that merges multiple :code:`ColumnParallelLinear` operators. Typically used for the first FFN layer with weighted activation functions (e.g., SiLU). This class handles the sharded weight loading logic of multiple weight matrices.
* :code:`QKVParallelLinear`: Parallel linear layer for the query, key, and value projections of the multi-head and grouped-query attention mechanisms. When number of key/value heads are less than the world size, this class replicates the key/value heads properly. This class handles the weight loading and replication of the weight matrices.
Note that all the linear layers above take :code:`linear_method` as an input. vLLM will set this parameter according to different quantization schemes to support weight quantization.
4. Implement the weight loading logic
-------------------------------------
You now need to implement the :code:`load_weights` method in your :code:`*ForCausalLM` class.
This method should load the weights from the HuggingFace's checkpoint file and assign them to the corresponding layers in your model. Specifically, for :code:`MergedColumnParallelLinear` and :code:`QKVParallelLinear` layers, if the original model has separated weight matrices, you need to load the different parts separately.
5. Register your model
----------------------
Finally, register your :code:`*ForCausalLM` class to the :code:`_VLLM_MODELS` in `vllm/model_executor/models/registry.py <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/registry.py>`_.
6. Out-of-Tree Model Integration
--------------------------------
You can integrate a model without modifying the vLLM codebase. Steps 2, 3, and 4 are still required, but you can skip steps 1 and 5. Instead, write a plugin to register your model. For general introduction of the plugin system, see :ref:`plugin_system`.
To register the model, use the following code:
.. code-block:: python
from vllm import ModelRegistry
from your_code import YourModelForCausalLM
ModelRegistry.register_model("YourModelForCausalLM", YourModelForCausalLM)
If your model imports modules that initialize CUDA, consider lazy-importing it to avoid errors like :code:`RuntimeError: Cannot re-initialize CUDA in forked subprocess`:
.. code-block:: python
from vllm import ModelRegistry
ModelRegistry.register_model("YourModelForCausalLM", "your_code:YourModelForCausalLM")
.. important::
If your model is a multimodal model, ensure the model class implements the :class:`~vllm.model_executor.models.interfaces.SupportsMultiModal` interface.
Read more about that :ref:`here <enabling_multimodal_inputs>`.
.. note::
Although you can directly put these code snippets in your script using ``vllm.LLM``, the recommended way is to place these snippets in a vLLM plugin. This ensures compatibility with various vLLM features like distributed inference and the API server.

143
docs/source/models/enabling_multimodal_inputs.md Normal file
View File

@ -0,0 +1,143 @@
(enabling-multimodal-inputs)=
# Enabling Multimodal Inputs
This document walks you through the steps to extend a vLLM model so that it accepts [multi-modal inputs](#multimodal-inputs).
```{seealso}
[Adding a New Model](adding-a-new-model)
```
## 1. Update the base vLLM model
It is assumed that you have already implemented the model in vLLM according to [these steps](#adding-a-new-model).
Further update the model as follows:
- Implement the {class}`~vllm.model_executor.models.interfaces.SupportsMultiModal` interface.
```diff
+ from vllm.model_executor.models.interfaces import SupportsMultiModal
- class YourModelForImage2Seq(nn.Module):
+ class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
```
```{note}
The model class does not have to be named {code}`*ForCausalLM`.
Check out [the HuggingFace Transformers documentation](https://huggingface.co/docs/transformers/model_doc/auto#multimodal) for some examples.
```
- If you haven't already done so, reserve a keyword parameter in {meth}`~torch.nn.Module.forward`
for each input tensor that corresponds to a multi-modal input, as shown in the following example:
```diff
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
+ pixel_values: torch.Tensor,
) -> SamplerOutput:
```
## 2. Register input mappers
For each modality type that the model accepts as input, decorate the model class with {meth}`MULTIMODAL_REGISTRY.register_input_mapper <vllm.multimodal.MultiModalRegistry.register_input_mapper>`.
This decorator accepts a function that maps multi-modal inputs to the keyword arguments you have previously defined in {meth}`~torch.nn.Module.forward`.
```diff
from vllm.model_executor.models.interfaces import SupportsMultiModal
+ from vllm.multimodal import MULTIMODAL_REGISTRY
+ @MULTIMODAL_REGISTRY.register_image_input_mapper()
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
```
A default mapper is available for each modality in the core vLLM library. This input mapper will be used if you do not provide your own function.
```{seealso}
[Input Processing Pipeline](#input-processing-pipeline)
```
## 3. Register maximum number of multi-modal tokens
For each modality type that the model accepts as input, calculate the maximum possible number of tokens per data item
and register it via {meth}`INPUT_REGISTRY.register_dummy_data <vllm.inputs.registry.InputRegistry.register_max_multimodal_tokens>`.
```diff
from vllm.inputs import INPUT_REGISTRY
from vllm.model_executor.models.interfaces import SupportsMultiModal
from vllm.multimodal import MULTIMODAL_REGISTRY
@MULTIMODAL_REGISTRY.register_image_input_mapper()
+ @MULTIMODAL_REGISTRY.register_max_image_tokens(<your_calculation>)
@INPUT_REGISTRY.register_dummy_data(<your_dummy_data_factory>)
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
```
Here are some examples:
- Image inputs (static feature size): [LLaVA-1.5 Model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava.py)
- Image inputs (dynamic feature size): [LLaVA-NeXT Model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava_next.py)
```{seealso}
[Input Processing Pipeline](#input-processing-pipeline)
```
## 4. (Optional) Register dummy data
During startup, dummy data is passed to the vLLM model to allocate memory. This only consists of text input by default, which may not be applicable to multi-modal models.
In such cases, you can define your own dummy data by registering a factory method via {meth}`INPUT_REGISTRY.register_dummy_data <vllm.inputs.registry.InputRegistry.register_dummy_data>`.
```diff
from vllm.inputs import INPUT_REGISTRY
from vllm.model_executor.models.interfaces import SupportsMultiModal
from vllm.multimodal import MULTIMODAL_REGISTRY
@MULTIMODAL_REGISTRY.register_image_input_mapper()
@MULTIMODAL_REGISTRY.register_max_image_tokens(<your_calculation>)
+ @INPUT_REGISTRY.register_dummy_data(<your_dummy_data_factory>)
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
```
```{note}
The dummy data should have the maximum possible number of multi-modal tokens, as described in the previous step.
```
Here are some examples:
- Image inputs (static feature size): [LLaVA-1.5 Model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava.py)
- Image inputs (dynamic feature size): [LLaVA-NeXT Model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava_next.py)
```{seealso}
[Input Processing Pipeline](#input-processing-pipeline)
```
## 5. (Optional) Register input processor
Sometimes, there is a need to process inputs at the {class}`~vllm.LLMEngine` level before they are passed to the model executor.
This is often due to the fact that unlike implementations in HuggingFace Transformers, the reshaping and/or expansion of multi-modal embeddings needs to take place outside model's {meth}`~torch.nn.Module.forward` call.
You can register input processors via {meth}`INPUT_REGISTRY.register_input_processor <vllm.inputs.registry.InputRegistry.register_input_processor>`.
```diff
from vllm.inputs import INPUT_REGISTRY
from vllm.model_executor.models.interfaces import SupportsMultiModal
from vllm.multimodal import MULTIMODAL_REGISTRY
@MULTIMODAL_REGISTRY.register_image_input_mapper()
@MULTIMODAL_REGISTRY.register_max_image_tokens(<your_calculation>)
@INPUT_REGISTRY.register_dummy_data(<your_dummy_data_factory>)
+ @INPUT_REGISTRY.register_input_processor(<your_input_processor>)
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
```
A common use case of input processors is inserting placeholder tokens to leverage the vLLM framework for attention mask generation.
Here are some examples:
- Insert static number of image tokens: [LLaVA-1.5 Model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava.py)
- Insert dynamic number of image tokens: [LLaVA-NeXT Model](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava_next.py)
```{seealso}
[Input Processing Pipeline](#input-processing-pipeline)
```

147
docs/source/models/enabling_multimodal_inputs.rst
View File

@ -1,147 +0,0 @@
.. _enabling_multimodal_inputs:
Enabling Multimodal Inputs
==========================
This document walks you through the steps to extend a vLLM model so that it accepts :ref:`multi-modal inputs <multimodal_inputs>`.
.. seealso::
:ref:`adding_a_new_model`
1. Update the base vLLM model
-----------------------------
It is assumed that you have already implemented the model in vLLM according to :ref:`these steps <adding_a_new_model>`.
Further update the model as follows:
- Implement the :class:`~vllm.model_executor.models.interfaces.SupportsMultiModal` interface.
.. code-block:: diff
+ from vllm.model_executor.models.interfaces import SupportsMultiModal
- class YourModelForImage2Seq(nn.Module):
+ class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
.. note::
The model class does not have to be named :code:`*ForCausalLM`.
Check out `the HuggingFace Transformers documentation <https://huggingface.co/docs/transformers/model_doc/auto#multimodal>`__ for some examples.
- If you haven't already done so, reserve a keyword parameter in :meth:`~torch.nn.Module.forward`
for each input tensor that corresponds to a multi-modal input, as shown in the following example:
.. code-block:: diff
def forward(
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
kv_caches: List[torch.Tensor],
attn_metadata: AttentionMetadata,
+ pixel_values: torch.Tensor,
) -> SamplerOutput:
2. Register input mappers
-------------------------
For each modality type that the model accepts as input, decorate the model class with :meth:`MULTIMODAL_REGISTRY.register_input_mapper <vllm.multimodal.MultiModalRegistry.register_input_mapper>`.
This decorator accepts a function that maps multi-modal inputs to the keyword arguments you have previously defined in :meth:`~torch.nn.Module.forward`.
.. code-block:: diff
from vllm.model_executor.models.interfaces import SupportsMultiModal
+ from vllm.multimodal import MULTIMODAL_REGISTRY
+ @MULTIMODAL_REGISTRY.register_image_input_mapper()
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
A default mapper is available for each modality in the core vLLM library. This input mapper will be used if you do not provide your own function.
.. seealso::
:ref:`input_processing_pipeline`
3. Register maximum number of multi-modal tokens
------------------------------------------------
For each modality type that the model accepts as input, calculate the maximum possible number of tokens per data item
and register it via :meth:`INPUT_REGISTRY.register_dummy_data <vllm.inputs.registry.InputRegistry.register_max_multimodal_tokens>`.
.. code-block:: diff
from vllm.inputs import INPUT_REGISTRY
from vllm.model_executor.models.interfaces import SupportsMultiModal
from vllm.multimodal import MULTIMODAL_REGISTRY
@MULTIMODAL_REGISTRY.register_image_input_mapper()
+ @MULTIMODAL_REGISTRY.register_max_image_tokens(<your_calculation>)
@INPUT_REGISTRY.register_dummy_data(<your_dummy_data_factory>)
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
Here are some examples:
- Image inputs (static feature size): `LLaVA-1.5 Model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava.py>`__
- Image inputs (dynamic feature size): `LLaVA-NeXT Model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava_next.py>`__
.. seealso::
:ref:`input_processing_pipeline`
4. (Optional) Register dummy data
---------------------------------
During startup, dummy data is passed to the vLLM model to allocate memory. This only consists of text input by default, which may not be applicable to multi-modal models.
In such cases, you can define your own dummy data by registering a factory method via :meth:`INPUT_REGISTRY.register_dummy_data <vllm.inputs.registry.InputRegistry.register_dummy_data>`.
.. code-block:: diff
from vllm.inputs import INPUT_REGISTRY
from vllm.model_executor.models.interfaces import SupportsMultiModal
from vllm.multimodal import MULTIMODAL_REGISTRY
@MULTIMODAL_REGISTRY.register_image_input_mapper()
@MULTIMODAL_REGISTRY.register_max_image_tokens(<your_calculation>)
+ @INPUT_REGISTRY.register_dummy_data(<your_dummy_data_factory>)
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
.. note::
The dummy data should have the maximum possible number of multi-modal tokens, as described in the previous step.
Here are some examples:
- Image inputs (static feature size): `LLaVA-1.5 Model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava.py>`__
- Image inputs (dynamic feature size): `LLaVA-NeXT Model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava_next.py>`__
.. seealso::
:ref:`input_processing_pipeline`
5. (Optional) Register input processor
--------------------------------------
Sometimes, there is a need to process inputs at the :class:`~vllm.LLMEngine` level before they are passed to the model executor.
This is often due to the fact that unlike implementations in HuggingFace Transformers, the reshaping and/or expansion of multi-modal embeddings needs to take place outside model's :meth:`~torch.nn.Module.forward` call.
You can register input processors via :meth:`INPUT_REGISTRY.register_input_processor <vllm.inputs.registry.InputRegistry.register_input_processor>`.
.. code-block:: diff
from vllm.inputs import INPUT_REGISTRY
from vllm.model_executor.models.interfaces import SupportsMultiModal
from vllm.multimodal import MULTIMODAL_REGISTRY
@MULTIMODAL_REGISTRY.register_image_input_mapper()
@MULTIMODAL_REGISTRY.register_max_image_tokens(<your_calculation>)
@INPUT_REGISTRY.register_dummy_data(<your_dummy_data_factory>)
+ @INPUT_REGISTRY.register_input_processor(<your_input_processor>)
class YourModelForImage2Seq(nn.Module, SupportsMultiModal):
A common use case of input processors is inserting placeholder tokens to leverage the vLLM framework for attention mask generation.
Here are some examples:
- Insert static number of image tokens: `LLaVA-1.5 Model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava.py>`__
- Insert dynamic number of image tokens: `LLaVA-NeXT Model <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/llava_next.py>`__
.. seealso::
:ref:`input_processing_pipeline`

138
docs/source/models/generative_models.md Normal file
View File

@ -0,0 +1,138 @@
(generative-models)=
# Generative Models
vLLM provides first-class support for generative models, which covers most of LLMs.
In vLLM, generative models implement the {class}`~vllm.model_executor.models.VllmModelForTextGeneration` interface.
Based on the final hidden states of the input, these models output log probabilities of the tokens to generate,
which are then passed through {class}`~vllm.model_executor.layers.Sampler` to obtain the final text.
## Offline Inference
The {class}`~vllm.LLM` class provides various methods for offline inference.
See [Engine Arguments](#engine-args) for a list of options when initializing the model.
For generative models, the only supported {code}`task` option is {code}`"generate"`.
Usually, this is automatically inferred so you don't have to specify it.
### `LLM.generate`
The {class}`~vllm.LLM.generate` method is available to all generative models in vLLM.
It is similar to [its counterpart in HF Transformers](https://huggingface.co/docs/transformers/main/en/main_classes/text_generation#transformers.GenerationMixin.generate),
except that tokenization and detokenization are also performed automatically.
```python
llm = LLM(model="facebook/opt-125m")
outputs = llm.generate("Hello, my name is")
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
You can optionally control the language generation by passing {class}`~vllm.SamplingParams`.
For example, you can use greedy sampling by setting {code}`temperature=0`:
```python
llm = LLM(model="facebook/opt-125m")
params = SamplingParams(temperature=0)
outputs = llm.generate("Hello, my name is", params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
A code example can be found in [examples/offline_inference.py](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference.py).
### `LLM.beam_search`
The {class}`~vllm.LLM.beam_search` method implements [beam search](https://huggingface.co/docs/transformers/en/generation_strategies#beam-search-decoding) on top of {class}`~vllm.LLM.generate`.
For example, to search using 5 beams and output at most 50 tokens:
```python
llm = LLM(model="facebook/opt-125m")
params = BeamSearchParams(beam_width=5, max_tokens=50)
outputs = llm.generate("Hello, my name is", params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
### `LLM.chat`
The {class}`~vllm.LLM.chat` method implements chat functionality on top of {class}`~vllm.LLM.generate`.
In particular, it accepts input similar to [OpenAI Chat Completions API](https://platform.openai.com/docs/api-reference/chat)
and automatically applies the model's [chat template](https://huggingface.co/docs/transformers/en/chat_templating) to format the prompt.
```{important}
In general, only instruction-tuned models have a chat template.
Base models may perform poorly as they are not trained to respond to the chat conversation.
```
```python
llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
conversation = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": "Hello"
},
{
"role": "assistant",
"content": "Hello! How can I assist you today?"
},
{
"role": "user",
"content": "Write an essay about the importance of higher education.",
},
]
outputs = llm.chat(conversation)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
A code example can be found in [examples/offline_inference_chat.py](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_chat.py).
If the model doesn't have a chat template or you want to specify another one,
you can explicitly pass a chat template:
```python
from vllm.entrypoints.chat_utils import load_chat_template
# You can find a list of existing chat templates under `examples/`
custom_template = load_chat_template(chat_template="<path_to_template>")
print("Loaded chat template:", custom_template)
outputs = llm.chat(conversation, chat_template=custom_template)
```
## Online Inference
Our [OpenAI Compatible Server](../serving/openai_compatible_server) can be used for online inference.
Please click on the above link for more details on how to launch the server.
### Completions API
Our Completions API is similar to `LLM.generate` but only accepts text.
It is compatible with [OpenAI Completions API](https://platform.openai.com/docs/api-reference/completions)
so that you can use OpenAI client to interact with it.
A code example can be found in [examples/openai_completion_client.py](https://github.com/vllm-project/vllm/blob/main/examples/openai_completion_client.py).
### Chat API
Our Chat API is similar to `LLM.chat`, accepting both text and [multi-modal inputs](#multimodal-inputs).
It is compatible with [OpenAI Chat Completions API](https://platform.openai.com/docs/api-reference/chat)
so that you can use OpenAI client to interact with it.
A code example can be found in [examples/openai_chat_completion_client.py](https://github.com/vllm-project/vllm/blob/main/examples/openai_chat_completion_client.py).

146
docs/source/models/generative_models.rst
View File

@ -1,146 +0,0 @@
.. _generative_models:
Generative Models
=================
vLLM provides first-class support for generative models, which covers most of LLMs.
In vLLM, generative models implement the :class:`~vllm.model_executor.models.VllmModelForTextGeneration` interface.
Based on the final hidden states of the input, these models output log probabilities of the tokens to generate,
which are then passed through :class:`~vllm.model_executor.layers.Sampler` to obtain the final text.
Offline Inference
-----------------
The :class:`~vllm.LLM` class provides various methods for offline inference.
See :ref:`Engine Arguments <engine_args>` for a list of options when initializing the model.
For generative models, the only supported :code:`task` option is :code:`"generate"`.
Usually, this is automatically inferred so you don't have to specify it.
``LLM.generate``
^^^^^^^^^^^^^^^^
The :class:`~vllm.LLM.generate` method is available to all generative models in vLLM.
It is similar to `its counterpart in HF Transformers <https://huggingface.co/docs/transformers/main/en/main_classes/text_generation#transformers.GenerationMixin.generate>`__,
except that tokenization and detokenization are also performed automatically.
.. code-block:: python
llm = LLM(model="facebook/opt-125m")
outputs = llm.generate("Hello, my name is")
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
You can optionally control the language generation by passing :class:`~vllm.SamplingParams`.
For example, you can use greedy sampling by setting :code:`temperature=0`:
.. code-block:: python
llm = LLM(model="facebook/opt-125m")
params = SamplingParams(temperature=0)
outputs = llm.generate("Hello, my name is", params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
A code example can be found in `examples/offline_inference.py <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference.py>`_.
``LLM.beam_search``
^^^^^^^^^^^^^^^^^^^
The :class:`~vllm.LLM.beam_search` method implements `beam search <https://huggingface.co/docs/transformers/en/generation_strategies#beam-search-decoding>`__ on top of :class:`~vllm.LLM.generate`.
For example, to search using 5 beams and output at most 50 tokens:
.. code-block:: python
llm = LLM(model="facebook/opt-125m")
params = BeamSearchParams(beam_width=5, max_tokens=50)
outputs = llm.generate("Hello, my name is", params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
``LLM.chat``
^^^^^^^^^^^^
The :class:`~vllm.LLM.chat` method implements chat functionality on top of :class:`~vllm.LLM.generate`.
In particular, it accepts input similar to `OpenAI Chat Completions API <https://platform.openai.com/docs/api-reference/chat>`__
and automatically applies the model's `chat template <https://huggingface.co/docs/transformers/en/chat_templating>`__ to format the prompt.
.. important::
In general, only instruction-tuned models have a chat template.
Base models may perform poorly as they are not trained to respond to the chat conversation.
.. code-block:: python
llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
conversation = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": "Hello"
},
{
"role": "assistant",
"content": "Hello! How can I assist you today?"
},
{
"role": "user",
"content": "Write an essay about the importance of higher education.",
},
]
outputs = llm.chat(conversation)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
A code example can be found in `examples/offline_inference_chat.py <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_chat.py>`_.
If the model doesn't have a chat template or you want to specify another one,
you can explicitly pass a chat template:
.. code-block:: python
from vllm.entrypoints.chat_utils import load_chat_template
# You can find a list of existing chat templates under `examples/`
custom_template = load_chat_template(chat_template="<path_to_template>")
print("Loaded chat template:", custom_template)
outputs = llm.chat(conversation, chat_template=custom_template)
Online Inference
----------------
Our `OpenAI Compatible Server <../serving/openai_compatible_server>`__ can be used for online inference.
Please click on the above link for more details on how to launch the server.
Completions API
^^^^^^^^^^^^^^^
Our Completions API is similar to ``LLM.generate`` but only accepts text.
It is compatible with `OpenAI Completions API <https://platform.openai.com/docs/api-reference/completions>`__
so that you can use OpenAI client to interact with it.
A code example can be found in `examples/openai_completion_client.py <https://github.com/vllm-project/vllm/blob/main/examples/openai_completion_client.py>`_.
Chat API
^^^^^^^^
Our Chat API is similar to ``LLM.chat``, accepting both text and :ref:`multi-modal inputs <multimodal_inputs>`.
It is compatible with `OpenAI Chat Completions API <https://platform.openai.com/docs/api-reference/chat>`__
so that you can use OpenAI client to interact with it.
A code example can be found in `examples/openai_chat_completion_client.py <https://github.com/vllm-project/vllm/blob/main/examples/openai_chat_completion_client.py>`_.

127
docs/source/models/pooling_models.md Normal file
View File

@ -0,0 +1,127 @@
(pooling-models)=
# Pooling Models
vLLM also supports pooling models, including embedding, reranking and reward models.
In vLLM, pooling models implement the {class}`~vllm.model_executor.models.VllmModelForPooling` interface.
These models use a {class}`~vllm.model_executor.layers.Pooler` to extract the final hidden states of the input
before returning them.
```{note}
We currently support pooling models primarily as a matter of convenience.
As shown in the [Compatibility Matrix](#compatibility-matrix), most vLLM features are not applicable to
pooling models as they only work on the generation or decode stage, so performance may not improve as much.
```
## Offline Inference
The {class}`~vllm.LLM` class provides various methods for offline inference.
See [Engine Arguments](#engine-args) for a list of options when initializing the model.
For pooling models, we support the following {code}`task` options:
- Embedding ({code}`"embed"` / {code}`"embedding"`)
- Classification ({code}`"classify"`)
- Sentence Pair Scoring ({code}`"score"`)
- Reward Modeling ({code}`"reward"`)
The selected task determines the default {class}`~vllm.model_executor.layers.Pooler` that is used:
- Embedding: Extract only the hidden states corresponding to the last token, and apply normalization.
- Classification: Extract only the hidden states corresponding to the last token, and apply softmax.
- Sentence Pair Scoring: Extract only the hidden states corresponding to the last token, and apply softmax.
- Reward Modeling: Extract all of the hidden states and return them directly.
When loading [Sentence Transformers](https://huggingface.co/sentence-transformers) models,
we attempt to override the default pooler based on its Sentence Transformers configuration file ({code}`modules.json`).
You can customize the model's pooling method via the {code}`override_pooler_config` option,
which takes priority over both the model's and Sentence Transformers's defaults.
### `LLM.encode`
The {class}`~vllm.LLM.encode` method is available to all pooling models in vLLM.
It returns the extracted hidden states directly, which is useful for reward models.
```python
llm = LLM(model="Qwen/Qwen2.5-Math-RM-72B", task="reward")
(output,) = llm.encode("Hello, my name is")
data = output.outputs.data
print(f"Data: {data!r}")
```
### `LLM.embed`
The {class}`~vllm.LLM.embed` method outputs an embedding vector for each prompt.
It is primarily designed for embedding models.
```python
llm = LLM(model="intfloat/e5-mistral-7b-instruct", task="embed")
(output,) = llm.embed("Hello, my name is")
embeds = output.outputs.embedding
print(f"Embeddings: {embeds!r} (size={len(embeds)})")
```
A code example can be found in [examples/offline_inference_embedding.py](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_embedding.py).
### `LLM.classify`
The {class}`~vllm.LLM.classify` method outputs a probability vector for each prompt.
It is primarily designed for classification models.
```python
llm = LLM(model="jason9693/Qwen2.5-1.5B-apeach", task="classify")
(output,) = llm.classify("Hello, my name is")
probs = output.outputs.probs
print(f"Class Probabilities: {probs!r} (size={len(probs)})")
```
A code example can be found in [examples/offline_inference_classification.py](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_classification.py).
### `LLM.score`
The {class}`~vllm.LLM.score` method outputs similarity scores between sentence pairs.
It is primarily designed for [cross-encoder models](https://www.sbert.net/examples/applications/cross-encoder/README.html).
These types of models serve as rerankers between candidate query-document pairs in RAG systems.
```{note}
vLLM can only perform the model inference component (e.g. embedding, reranking) of RAG.
To handle RAG at a higher level, you should use integration frameworks such as [LangChain](https://github.com/langchain-ai/langchain).
```
```python
llm = LLM(model="BAAI/bge-reranker-v2-m3", task="score")
(output,) = llm.score("What is the capital of France?",
"The capital of Brazil is Brasilia.")
score = output.outputs.score
print(f"Score: {score}")
```
A code example can be found in [examples/offline_inference_scoring.py](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_scoring.py).
## Online Inference
Our [OpenAI Compatible Server](../serving/openai_compatible_server.md) can be used for online inference.
Please click on the above link for more details on how to launch the server.
### Embeddings API
Our Embeddings API is similar to `LLM.embed`, accepting both text and [multi-modal inputs](#multimodal-inputs).
The text-only API is compatible with [OpenAI Embeddings API](https://platform.openai.com/docs/api-reference/embeddings)
so that you can use OpenAI client to interact with it.
A code example can be found in [examples/openai_embedding_client.py](https://github.com/vllm-project/vllm/blob/main/examples/openai_embedding_client.py).
The multi-modal API is an extension of the [OpenAI Embeddings API](https://platform.openai.com/docs/api-reference/embeddings)
that incorporates [OpenAI Chat Completions API](https://platform.openai.com/docs/api-reference/chat),
so it is not part of the OpenAI standard. Please see [](#multimodal-inputs) for more details on how to use it.
### Score API
Our Score API is similar to `LLM.score`.
Please see [this page](#score-api) for more details on how to use it.

136
docs/source/models/pooling_models.rst
View File

@ -1,136 +0,0 @@
.. _pooling_models:
Pooling Models
==============
vLLM also supports pooling models, including embedding, reranking and reward models.
In vLLM, pooling models implement the :class:`~vllm.model_executor.models.VllmModelForPooling` interface.
These models use a :class:`~vllm.model_executor.layers.Pooler` to extract the final hidden states of the input
before returning them.
.. note::
We currently support pooling models primarily as a matter of convenience.
As shown in the :ref:`Compatibility Matrix <compatibility_matrix>`, most vLLM features are not applicable to
pooling models as they only work on the generation or decode stage, so performance may not improve as much.
Offline Inference
-----------------
The :class:`~vllm.LLM` class provides various methods for offline inference.
See :ref:`Engine Arguments <engine_args>` for a list of options when initializing the model.
For pooling models, we support the following :code:`task` options:
- Embedding (:code:`"embed"` / :code:`"embedding"`)
- Classification (:code:`"classify"`)
- Sentence Pair Scoring (:code:`"score"`)
- Reward Modeling (:code:`"reward"`)
The selected task determines the default :class:`~vllm.model_executor.layers.Pooler` that is used:
- Embedding: Extract only the hidden states corresponding to the last token, and apply normalization.
- Classification: Extract only the hidden states corresponding to the last token, and apply softmax.
- Sentence Pair Scoring: Extract only the hidden states corresponding to the last token, and apply softmax.
- Reward Modeling: Extract all of the hidden states and return them directly.
When loading `Sentence Transformers <https://huggingface.co/sentence-transformers>`__ models,
we attempt to override the default pooler based on its Sentence Transformers configuration file (:code:`modules.json`).
You can customize the model's pooling method via the :code:`override_pooler_config` option,
which takes priority over both the model's and Sentence Transformers's defaults.
``LLM.encode``
^^^^^^^^^^^^^^
The :class:`~vllm.LLM.encode` method is available to all pooling models in vLLM.
It returns the extracted hidden states directly, which is useful for reward models.
.. code-block:: python
llm = LLM(model="Qwen/Qwen2.5-Math-RM-72B", task="reward")
(output,) = llm.encode("Hello, my name is")
data = output.outputs.data
print(f"Data: {data!r}")
``LLM.embed``
^^^^^^^^^^^^^
The :class:`~vllm.LLM.embed` method outputs an embedding vector for each prompt.
It is primarily designed for embedding models.
.. code-block:: python
llm = LLM(model="intfloat/e5-mistral-7b-instruct", task="embed")
(output,) = llm.embed("Hello, my name is")
embeds = output.outputs.embedding
print(f"Embeddings: {embeds!r} (size={len(embeds)})")
A code example can be found in `examples/offline_inference_embedding.py <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_embedding.py>`_.
``LLM.classify``
^^^^^^^^^^^^^^^^
The :class:`~vllm.LLM.classify` method outputs a probability vector for each prompt.
It is primarily designed for classification models.
.. code-block:: python
llm = LLM(model="jason9693/Qwen2.5-1.5B-apeach", task="classify")
(output,) = llm.classify("Hello, my name is")
probs = output.outputs.probs
print(f"Class Probabilities: {probs!r} (size={len(probs)})")
A code example can be found in `examples/offline_inference_classification.py <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_classification.py>`_.
``LLM.score``
^^^^^^^^^^^^^
The :class:`~vllm.LLM.score` method outputs similarity scores between sentence pairs.
It is primarily designed for `cross-encoder models <https://www.sbert.net/examples/applications/cross-encoder/README.html>`__.
These types of models serve as rerankers between candidate query-document pairs in RAG systems.
.. note::
vLLM can only perform the model inference component (e.g. embedding, reranking) of RAG.
To handle RAG at a higher level, you should use integration frameworks such as `LangChain <https://github.com/langchain-ai/langchain>`_.
.. code-block:: python
llm = LLM(model="BAAI/bge-reranker-v2-m3", task="score")
(output,) = llm.score("What is the capital of France?",
"The capital of Brazil is Brasilia.")
score = output.outputs.score
print(f"Score: {score}")
A code example can be found in `examples/offline_inference_scoring.py <https://github.com/vllm-project/vllm/blob/main/examples/offline_inference_scoring.py>`_.
Online Inference
----------------
Our `OpenAI Compatible Server <../serving/openai_compatible_server>`__ can be used for online inference.
Please click on the above link for more details on how to launch the server.
Embeddings API
^^^^^^^^^^^^^^
Our Embeddings API is similar to ``LLM.embed``, accepting both text and :ref:`multi-modal inputs <multimodal_inputs>`.
The text-only API is compatible with `OpenAI Embeddings API <https://platform.openai.com/docs/api-reference/embeddings>`__
so that you can use OpenAI client to interact with it.
A code example can be found in `examples/openai_embedding_client.py <https://github.com/vllm-project/vllm/blob/main/examples/openai_embedding_client.py>`_.
The multi-modal API is an extension of the `OpenAI Embeddings API <https://platform.openai.com/docs/api-reference/embeddings>`__
that incorporates `OpenAI Chat Completions API <https://platform.openai.com/docs/api-reference/chat>`__,
so it is not part of the OpenAI standard. Please see :ref:`this page <multimodal_inputs>` for more details on how to use it.
Score API
^^^^^^^^^
Our Score API is similar to ``LLM.score``.
Please see `this page <../serving/openai_compatible_server.html#score-api-for-cross-encoder-models>`__ for more details on how to use it.

319
docs/source/models/supported_models.rst → docs/source/models/supported_models.md
View File

@ -1,84 +1,78 @@
.. _supported_models:
(supported-models)=
Supported Models
================
# Supported Models
vLLM supports generative and pooling models across various tasks.
If a model supports more than one task, you can set the task via the :code:`--task` argument.
If a model supports more than one task, you can set the task via the {code}`--task` argument.
For each task, we list the model architectures that have been implemented in vLLM.
Alongside each architecture, we include some popular models that use it.
Loading a Model
^^^^^^^^^^^^^^^
## Loading a Model
HuggingFace Hub
+++++++++++++++
### HuggingFace Hub
By default, vLLM loads models from `HuggingFace (HF) Hub <https://huggingface.co/models>`_.
By default, vLLM loads models from [HuggingFace (HF) Hub](https://huggingface.co/models).
To determine whether a given model is supported, you can check the :code:`config.json` file inside the HF repository.
If the :code:`"architectures"` field contains a model architecture listed below, then it should be supported in theory.
To determine whether a given model is supported, you can check the {code}`config.json` file inside the HF repository.
If the {code}`"architectures"` field contains a model architecture listed below, then it should be supported in theory.
.. tip::
The easiest way to check if your model is really supported at runtime is to run the program below:
````{tip}
The easiest way to check if your model is really supported at runtime is to run the program below:
.. code-block:: python
```python
from vllm import LLM
from vllm import LLM
# For generative models (task=generate) only
llm = LLM(model=..., task="generate") # Name or path of your model
output = llm.generate("Hello, my name is")
print(output)
# For generative models (task=generate) only
llm = LLM(model=..., task="generate") # Name or path of your model
output = llm.generate("Hello, my name is")
print(output)
# For pooling models (task={embed,classify,reward}) only
llm = LLM(model=..., task="embed") # Name or path of your model
output = llm.encode("Hello, my name is")
print(output)
```
# For pooling models (task={embed,classify,reward}) only
llm = LLM(model=..., task="embed") # Name or path of your model
output = llm.encode("Hello, my name is")
print(output)
If vLLM successfully returns text (for generative models) or hidden states (for pooling models), it indicates that your model is supported.
````
If vLLM successfully returns text (for generative models) or hidden states (for pooling models), it indicates that your model is supported.
Otherwise, please refer to [Adding a New Model](#adding-a-new-model) and [Enabling Multimodal Inputs](#enabling-multimodal-inputs) for instructions on how to implement your model in vLLM.
Alternatively, you can [open an issue on GitHub](https://github.com/vllm-project/vllm/issues/new/choose) to request vLLM support.
Otherwise, please refer to :ref:`Adding a New Model <adding_a_new_model>` and :ref:`Enabling Multimodal Inputs <enabling_multimodal_inputs>`
for instructions on how to implement your model in vLLM.
Alternatively, you can `open an issue on GitHub <https://github.com/vllm-project/vllm/issues/new/choose>`_ to request vLLM support.
### ModelScope
ModelScope
++++++++++
To use models from [ModelScope](https://www.modelscope.cn) instead of HuggingFace Hub, set an environment variable:
To use models from `ModelScope <https://www.modelscope.cn>`_ instead of HuggingFace Hub, set an environment variable:
```shell
$ export VLLM_USE_MODELSCOPE=True
```
.. code-block:: shell
And use with {code}`trust_remote_code=True`.
$ export VLLM_USE_MODELSCOPE=True
```python
from vllm import LLM
And use with :code:`trust_remote_code=True`.
llm = LLM(model=..., revision=..., task=..., trust_remote_code=True)
.. code-block:: python
# For generative models (task=generate) only
output = llm.generate("Hello, my name is")
print(output)
from vllm import LLM
# For pooling models (task={embed,classify,reward}) only
output = llm.encode("Hello, my name is")
print(output)
```
llm = LLM(model=..., revision=..., task=..., trust_remote_code=True)
## List of Text-only Language Models
# For generative models (task=generate) only
output = llm.generate("Hello, my name is")
print(output)
### Generative Models
# For pooling models (task={embed,classify,reward}) only
output = llm.encode("Hello, my name is")
print(output)
See [this page](#generative-models) for more information on how to use generative models.
List of Text-only Language Models
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Generative Models
+++++++++++++++++
See :ref:`this page <generative_models>` for more information on how to use generative models.
Text Generation (``--task generate``)
-------------------------------------
#### Text Generation (`--task generate`)
```{eval-rst}
.. list-table::
:widths: 25 25 50 5 5
:header-rows: 1
@ -86,8 +80,8 @@ Text Generation (``--task generate``)
* - Architecture
- Models
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
* - :code:`AquilaForCausalLM`
- Aquila, Aquila2
- :code:`BAAI/Aquila-7B`, :code:`BAAI/AquilaChat-7B`, etc.
@ -358,29 +352,32 @@ Text Generation (``--task generate``)
- :code:`xverse/XVERSE-7B-Chat`, :code:`xverse/XVERSE-13B-Chat`, :code:`xverse/XVERSE-65B-Chat`, etc.
- ✅︎
- ✅︎
```
.. note::
Currently, the ROCm version of vLLM supports Mistral and Mixtral only for context lengths up to 4096.
```{note}
Currently, the ROCm version of vLLM supports Mistral and Mixtral only for context lengths up to 4096.
```
Pooling Models
++++++++++++++
### Pooling Models
See :ref:`this page <pooling_models>` for more information on how to use pooling models.
See [this page](pooling-models) for more information on how to use pooling models.
.. important::
Since some model architectures support both generative and pooling tasks,
you should explicitly specify the task type to ensure that the model is used in pooling mode instead of generative mode.
```{important}
Since some model architectures support both generative and pooling tasks,
you should explicitly specify the task type to ensure that the model is used in pooling mode instead of generative mode.
```
Text Embedding (``--task embed``)
---------------------------------
#### Text Embedding (`--task embed`)
Any text generation model can be converted into an embedding model by passing :code:`--task embed`.
Any text generation model can be converted into an embedding model by passing {code}`--task embed`.
.. note::
To get the best results, you should use pooling models that are specifically trained as such.
```{note}
To get the best results, you should use pooling models that are specifically trained as such.
```
The following table lists those that are tested in vLLM.
```{eval-rst}
.. list-table::
:widths: 25 25 50 5 5
:header-rows: 1
@ -388,8 +385,8 @@ The following table lists those that are tested in vLLM.
* - Architecture
- Models
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
* - :code:`BertModel`
- BERT-based
- :code:`BAAI/bge-base-en-v1.5`, etc.
@ -425,21 +422,24 @@ The following table lists those that are tested in vLLM.
- :code:`intfloat/multilingual-e5-large`, etc.
-
-
```
.. note::
:code:`ssmits/Qwen2-7B-Instruct-embed-base` has an improperly defined Sentence Transformers config.
You should manually set mean pooling by passing :code:`--override-pooler-config '{"pooling_type": "MEAN"}'`.
```{note}
{code}`ssmits/Qwen2-7B-Instruct-embed-base` has an improperly defined Sentence Transformers config.
You should manually set mean pooling by passing {code}`--override-pooler-config '{"pooling_type": "MEAN"}'`.
```
.. note::
Unlike base Qwen2, :code:`Alibaba-NLP/gte-Qwen2-7B-instruct` uses bi-directional attention.
You can set :code:`--hf-overrides '{"is_causal": false}'` to change the attention mask accordingly.
```{note}
Unlike base Qwen2, {code}`Alibaba-NLP/gte-Qwen2-7B-instruct` uses bi-directional attention.
You can set {code}`--hf-overrides '{"is_causal": false}'` to change the attention mask accordingly.
On the other hand, its 1.5B variant (:code:`Alibaba-NLP/gte-Qwen2-1.5B-instruct`) uses causal attention
despite being described otherwise on its model card.
On the other hand, its 1.5B variant ({code}`Alibaba-NLP/gte-Qwen2-1.5B-instruct`) uses causal attention
despite being described otherwise on its model card.
```
Reward Modeling (``--task reward``)
-----------------------------------
#### Reward Modeling (`--task reward`)
```{eval-rst}
.. list-table::
:widths: 25 25 50 5 5
:header-rows: 1
@ -447,8 +447,8 @@ Reward Modeling (``--task reward``)
* - Architecture
- Models
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
* - :code:`LlamaForCausalLM`
- Llama-based
- :code:`peiyi9979/math-shepherd-mistral-7b-prm`, etc.
@ -459,14 +459,16 @@ Reward Modeling (``--task reward``)
- :code:`Qwen/Qwen2.5-Math-RM-72B`, etc.
- ✅︎
- ✅︎
```
.. important::
For process-supervised reward models such as :code:`peiyi9979/math-shepherd-mistral-7b-prm`, the pooling config should be set explicitly,
e.g.: :code:`--override-pooler-config '{"pooling_type": "STEP", "step_tag_id": 123, "returned_token_ids": [456, 789]}'`.
```{important}
For process-supervised reward models such as {code}`peiyi9979/math-shepherd-mistral-7b-prm`, the pooling config should be set explicitly,
e.g.: {code}`--override-pooler-config '{"pooling_type": "STEP", "step_tag_id": 123, "returned_token_ids": [456, 789]}'`.
```
Classification (``--task classify``)
------------------------------------
#### Classification (`--task classify`)
```{eval-rst}
.. list-table::
:widths: 25 25 50 5 5
:header-rows: 1
@ -474,8 +476,8 @@ Classification (``--task classify``)
* - Architecture
- Models
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
* - :code:`JambaForSequenceClassification`
- Jamba
- :code:`ai21labs/Jamba-tiny-reward-dev`, etc.
@ -486,10 +488,11 @@ Classification (``--task classify``)
- :code:`jason9693/Qwen2.5-1.5B-apeach`, etc.
- ✅︎
- ✅︎
```
Sentence Pair Scoring (``--task score``)
----------------------------------------
#### Sentence Pair Scoring (`--task score`)
```{eval-rst}
.. list-table::
:widths: 25 25 50 5 5
:header-rows: 1
@ -497,8 +500,8 @@ Sentence Pair Scoring (``--task score``)
* - Architecture
- Models
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
* - :code:`BertForSequenceClassification`
- BERT-based
- :code:`cross-encoder/ms-marco-MiniLM-L-6-v2`, etc.
@ -514,37 +517,36 @@ Sentence Pair Scoring (``--task score``)
- :code:`BAAI/bge-reranker-v2-m3`, etc.
-
-
```
.. _supported_mm_models:
(supported-mm-models)=
List of Multimodal Language Models
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
## List of Multimodal Language Models
The following modalities are supported depending on the model:
- **T**\ ext
- **I**\ mage
- **V**\ ideo
- **A**\ udio
- **T**ext
- **I**mage
- **V**ideo
- **A**udio
Any combination of modalities joined by :code:`+` are supported.
Any combination of modalities joined by {code}`+` are supported.
- e.g.: :code:`T + I` means that the model supports text-only, image-only, and text-with-image inputs.
- e.g.: {code}`T + I` means that the model supports text-only, image-only, and text-with-image inputs.
On the other hand, modalities separated by :code:`/` are mutually exclusive.
On the other hand, modalities separated by {code}`/` are mutually exclusive.
- e.g.: :code:`T / I` means that the model supports text-only and image-only inputs, but not text-with-image inputs.
- e.g.: {code}`T / I` means that the model supports text-only and image-only inputs, but not text-with-image inputs.
See :ref:`this page <multimodal_inputs>` on how to pass multi-modal inputs to the model.
See [this page](#multimodal-inputs) on how to pass multi-modal inputs to the model.
Generative Models
+++++++++++++++++
### Generative Models
See :ref:`this page <generative_models>` for more information on how to use generative models.
See [this page](#generative-models) for more information on how to use generative models.
Text Generation (``--task generate``)
-------------------------------------
#### Text Generation (`--task generate`)
```{eval-rst}
.. list-table::
:widths: 25 25 15 20 5 5 5
:header-rows: 1
@ -553,8 +555,8 @@ Text Generation (``--task generate``)
- Models
- Inputs
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
- V1
* - :code:`AriaForConditionalGeneration`
- Aria
@ -717,55 +719,64 @@ Text Generation (``--task generate``)
-
- ✅︎
-
```
| :sup:`E` Pre-computed embeddings can be inputted for this modality.
| :sup:`+` Multiple items can be inputted per text prompt for this modality.
```{eval-rst}
:sup:`E` Pre-computed embeddings can be inputted for this modality.
.. important::
To enable multiple multi-modal items per text prompt, you have to set :code:`limit_mm_per_prompt` (offline inference)
or :code:`--limit-mm-per-prompt` (online inference). For example, to enable passing up to 4 images per text prompt:
:sup:`+` Multiple items can be inputted per text prompt for this modality.
```
.. code-block:: python
````{important}
To enable multiple multi-modal items per text prompt, you have to set {code}`limit_mm_per_prompt` (offline inference)
or {code}`--limit-mm-per-prompt` (online inference). For example, to enable passing up to 4 images per text prompt:
llm = LLM(
model="Qwen/Qwen2-VL-7B-Instruct",
limit_mm_per_prompt={"image": 4},
)
```python
llm = LLM(
model="Qwen/Qwen2-VL-7B-Instruct",
limit_mm_per_prompt={"image": 4},
)
```
.. code-block:: bash
```bash
vllm serve Qwen/Qwen2-VL-7B-Instruct --limit-mm-per-prompt image=4
```
````
vllm serve Qwen/Qwen2-VL-7B-Instruct --limit-mm-per-prompt image=4
```{note}
vLLM currently only supports adding LoRA to the language backbone of multimodal models.
```
.. note::
vLLM currently only supports adding LoRA to the language backbone of multimodal models.
```{note}
To use {code}`TIGER-Lab/Mantis-8B-siglip-llama3`, you have to install their GitHub repo ({code}`pip install git+https://github.com/TIGER-AI-Lab/Mantis.git`)
and pass {code}`--hf_overrides '{"architectures": ["MantisForConditionalGeneration"]}'` when running vLLM.
```
.. note::
To use :code:`TIGER-Lab/Mantis-8B-siglip-llama3`, you have to install their GitHub repo (:code:`pip install git+https://github.com/TIGER-AI-Lab/Mantis.git`)
and pass :code:`--hf_overrides '{"architectures": ["MantisForConditionalGeneration"]}'` when running vLLM.
```{note}
The official {code}`openbmb/MiniCPM-V-2` doesn't work yet, so we need to use a fork ({code}`HwwwH/MiniCPM-V-2`) for now.
For more details, please see: <https://github.com/vllm-project/vllm/pull/4087#issuecomment-2250397630>
```
.. note::
The official :code:`openbmb/MiniCPM-V-2` doesn't work yet, so we need to use a fork (:code:`HwwwH/MiniCPM-V-2`) for now.
For more details, please see: https://github.com/vllm-project/vllm/pull/4087#issuecomment-2250397630
### Pooling Models
Pooling Models
++++++++++++++
See [this page](pooling-models) for more information on how to use pooling models.
See :ref:`this page <pooling_models>` for more information on how to use pooling models.
```{important}
Since some model architectures support both generative and pooling tasks,
you should explicitly specify the task type to ensure that the model is used in pooling mode instead of generative mode.
```
.. important::
Since some model architectures support both generative and pooling tasks,
you should explicitly specify the task type to ensure that the model is used in pooling mode instead of generative mode.
#### Text Embedding (`--task embed`)
Text Embedding (``--task embed``)
---------------------------------
Any text generation model can be converted into an embedding model by passing {code}`--task embed`.
Any text generation model can be converted into an embedding model by passing :code:`--task embed`.
.. note::
To get the best results, you should use pooling models that are specifically trained as such.
```{note}
To get the best results, you should use pooling models that are specifically trained as such.
```
The following table lists those that are tested in vLLM.
```{eval-rst}
.. list-table::
:widths: 25 25 15 25 5 5
:header-rows: 1
@ -774,8 +785,8 @@ The following table lists those that are tested in vLLM.
- Models
- Inputs
- Example HF Models
- :ref:`LoRA <lora>`
- :ref:`PP <distributed_serving>`
- :ref:`LoRA <lora-adapter>`
- :ref:`PP <distributed-serving>`
* - :code:`LlavaNextForConditionalGeneration`
- LLaVA-NeXT-based
- T / I
@ -794,25 +805,23 @@ The following table lists those that are tested in vLLM.
- :code:`MrLight/dse-qwen2-2b-mrl-v1`
-
- ✅︎
```
----
______________________________________________________________________
Model Support Policy
=====================
# Model Support Policy
At vLLM, we are committed to facilitating the integration and support of third-party models within our ecosystem. Our approach is designed to balance the need for robustness and the practical limitations of supporting a wide range of models. Here’s how we manage third-party model support:
1. **Community-Driven Support**: We encourage community contributions for adding new models. When a user requests support for a new model, we welcome pull requests (PRs) from the community. These contributions are evaluated primarily on the sensibility of the output they generate, rather than strict consistency with existing implementations such as those in transformers. **Call for contribution:** PRs coming directly from model vendors are greatly appreciated!
2. **Best-Effort Consistency**: While we aim to maintain a level of consistency between the models implemented in vLLM and other frameworks like transformers, complete alignment is not always feasible. Factors like acceleration techniques and the use of low-precision computations can introduce discrepancies. Our commitment is to ensure that the implemented models are functional and produce sensible results.
.. tip::
When comparing the output of :code:`model.generate` from HuggingFace Transformers with the output of :code:`llm.generate` from vLLM, note that the former reads the model's generation config file (i.e., `generation_config.json <https://github.com/huggingface/transformers/blob/19dabe96362803fb0a9ae7073d03533966598b17/src/transformers/generation/utils.py#L1945>`__) and applies the default parameters for generation, while the latter only uses the parameters passed to the function. Ensure all sampling parameters are identical when comparing outputs.
```{tip}
When comparing the output of {code}`model.generate` from HuggingFace Transformers with the output of {code}`llm.generate` from vLLM, note that the former reads the model's generation config file (i.e., [generation_config.json](https://github.com/huggingface/transformers/blob/19dabe96362803fb0a9ae7073d03533966598b17/src/transformers/generation/utils.py#L1945)) and applies the default parameters for generation, while the latter only uses the parameters passed to the function. Ensure all sampling parameters are identical when comparing outputs.
```
3. **Issue Resolution and Model Updates**: Users are encouraged to report any bugs or issues they encounter with third-party models. Proposed fixes should be submitted via PRs, with a clear explanation of the problem and the rationale behind the proposed solution. If a fix for one model impacts another, we rely on the community to highlight and address these cross-model dependencies. Note: for bugfix PRs, it is good etiquette to inform the original author to seek their feedback.
4. **Monitoring and Updates**: Users interested in specific models should monitor the commit history for those models (e.g., by tracking changes in the main/vllm/model_executor/models directory). This proactive approach helps users stay informed about updates and changes that may affect the models they use.
5. **Selective Focus**: Our resources are primarily directed towards models with significant user interest and impact. Models that are less frequently used may receive less attention, and we rely on the community to play a more active role in their upkeep and improvement.
Through this approach, vLLM fosters a collaborative environment where both the core development team and the broader community contribute to the robustness and diversity of the third-party models supported in our ecosystem.
@ -821,7 +830,7 @@ Note that, as an inference engine, vLLM does not introduce new models. Therefore
We have the following levels of testing for models:
1. **Strict Consistency**: We compare the output of the model with the output of the model in the HuggingFace Transformers library under greedy decoding. This is the most stringent test. Please refer to `models tests <https://github.com/vllm-project/vllm/blob/main/tests/models>`_ for the models that have passed this test.
1. **Strict Consistency**: We compare the output of the model with the output of the model in the HuggingFace Transformers library under greedy decoding. This is the most stringent test. Please refer to [models tests](https://github.com/vllm-project/vllm/blob/main/tests/models) for the models that have passed this test.
2. **Output Sensibility**: We check if the output of the model is sensible and coherent, by measuring the perplexity of the output and checking for any obvious errors. This is a less stringent test.
3. **Runtime Functionality**: We check if the model can be loaded and run without errors. This is the least stringent test. Please refer to `functionality tests <https://github.com/vllm-project/vllm/tree/main/tests>`_ and `examples <https://github.com/vllm-project/vllm/tree/main/examples>`_ for the models that have passed this test.
3. **Runtime Functionality**: We check if the model can be loaded and run without errors. This is the least stringent test. Please refer to [functionality tests](https://github.com/vllm-project/vllm/tree/main/tests) and [examples](https://github.com/vllm-project/vllm/tree/main/examples) for the models that have passed this test.
4. **Community Feedback**: We rely on the community to provide feedback on the models. If a model is broken or not working as expected, we encourage users to raise issues to report it or open pull requests to fix it. The rest of the models fall under this category.

28
docs/source/performance/benchmarks.md Normal file
View File

@ -0,0 +1,28 @@
(benchmarks)=
# Benchmark Suites
vLLM contains two sets of benchmarks:
- [Performance benchmarks](#performance-benchmarks)
- [Nightly benchmarks](#nightly-benchmarks)
(performance-benchmarks)=
## Performance Benchmarks
The performance benchmarks are used for development to confirm whether new changes improve performance under various workloads. They are triggered on every commit with both the `perf-benchmarks` and `ready` labels, and when a PR is merged into vLLM.
The latest performance results are hosted on the public [vLLM Performance Dashboard](https://perf.vllm.ai).
More information on the performance benchmarks and their parameters can be found [here](https://github.com/vllm-project/vllm/blob/main/.buildkite/nightly-benchmarks/performance-benchmarks-descriptions.md).
(nightly-benchmarks)=
## Nightly Benchmarks
These compare vLLM's performance against alternatives (`tgi`, `trt-llm`, and `lmdeploy`) when there are major updates of vLLM (e.g., bumping up to a new version). They are primarily intended for consumers to evaluate when to choose vLLM over other options and are triggered on every commit with both the `perf-benchmarks` and `nightly-benchmarks` labels.
The latest nightly benchmark results are shared in major release blog posts such as [vLLM v0.6.0](https://blog.vllm.ai/2024/09/05/perf-update.html).
More information on the nightly benchmarks and their parameters can be found [here](https://github.com/vllm-project/vllm/blob/main/.buildkite/nightly-benchmarks/nightly-descriptions.md).

33
docs/source/performance/benchmarks.rst
View File

@ -1,33 +0,0 @@
.. _benchmarks:
================
Benchmark Suites
================
vLLM contains two sets of benchmarks:
+ :ref:`Performance benchmarks <performance_benchmarks>`
+ :ref:`Nightly benchmarks <nightly_benchmarks>`
.. _performance_benchmarks:
Performance Benchmarks
----------------------
The performance benchmarks are used for development to confirm whether new changes improve performance under various workloads. They are triggered on every commit with both the ``perf-benchmarks`` and ``ready`` labels, and when a PR is merged into vLLM.
The latest performance results are hosted on the public `vLLM Performance Dashboard <https://perf.vllm.ai>`_.
More information on the performance benchmarks and their parameters can be found `here <https://github.com/vllm-project/vllm/blob/main/.buildkite/nightly-benchmarks/performance-benchmarks-descriptions.md>`__.
.. _nightly_benchmarks:
Nightly Benchmarks
------------------
These compare vLLM's performance against alternatives (``tgi``, ``trt-llm``, and ``lmdeploy``) when there are major updates of vLLM (e.g., bumping up to a new version). They are primarily intended for consumers to evaluate when to choose vLLM over other options and are triggered on every commit with both the ``perf-benchmarks`` and ``nightly-benchmarks`` labels.
The latest nightly benchmark results are shared in major release blog posts such as `vLLM v0.6.0 <https://blog.vllm.ai/2024/09/05/perf-update.html>`_.
More information on the nightly benchmarks and their parameters can be found `here <https://github.com/vllm-project/vllm/blob/main/.buildkite/nightly-benchmarks/nightly-descriptions.md>`__.

78
docs/source/quantization/auto_awq.md Normal file
View File

@ -0,0 +1,78 @@
(auto-awq)=
# AutoAWQ
```{warning}
Please note that AWQ support in vLLM is under-optimized at the moment. We would recommend using the unquantized version of the model for better
accuracy and higher throughput. Currently, you can use AWQ as a way to reduce memory footprint. As of now, it is more suitable for low latency
inference with small number of concurrent requests. vLLM's AWQ implementation have lower throughput than unquantized version.
```
To create a new 4-bit quantized model, you can leverage [AutoAWQ](https://github.com/casper-hansen/AutoAWQ).
Quantizing reduces the model's precision from FP16 to INT4 which effectively reduces the file size by ~70%.
The main benefits are lower latency and memory usage.
You can quantize your own models by installing AutoAWQ or picking one of the [400+ models on Huggingface](https://huggingface.co/models?sort=trending&search=awq).
```console
$ pip install autoawq
```
After installing AutoAWQ, you are ready to quantize a model. Here is an example of how to quantize `mistralai/Mistral-7B-Instruct-v0.2`:
```python
from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer
model_path = 'mistralai/Mistral-7B-Instruct-v0.2'
quant_path = 'mistral-instruct-v0.2-awq'
quant_config = { "zero_point": True, "q_group_size": 128, "w_bit": 4, "version": "GEMM" }
# Load model
model = AutoAWQForCausalLM.from_pretrained(
model_path, **{"low_cpu_mem_usage": True, "use_cache": False}
)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
# Quantize
model.quantize(tokenizer, quant_config=quant_config)
# Save quantized model
model.save_quantized(quant_path)
tokenizer.save_pretrained(quant_path)
print(f'Model is quantized and saved at "{quant_path}"')
```
To run an AWQ model with vLLM, you can use [TheBloke/Llama-2-7b-Chat-AWQ](https://huggingface.co/TheBloke/Llama-2-7b-Chat-AWQ) with the following command:
```console
$ python examples/llm_engine_example.py --model TheBloke/Llama-2-7b-Chat-AWQ --quantization awq
```
AWQ models are also supported directly through the LLM entrypoint:
```python
from vllm import LLM, SamplingParams
# Sample prompts.
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Create an LLM.
llm = LLM(model="TheBloke/Llama-2-7b-Chat-AWQ", quantization="AWQ")
# Generate texts from the prompts. The output is a list of RequestOutput objects
# that contain the prompt, generated text, and other information.
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```

79
docs/source/quantization/auto_awq.rst
View File

@ -1,79 +0,0 @@
.. _auto_awq:
AutoAWQ
==================
.. warning::
Please note that AWQ support in vLLM is under-optimized at the moment. We would recommend using the unquantized version of the model for better
accuracy and higher throughput. Currently, you can use AWQ as a way to reduce memory footprint. As of now, it is more suitable for low latency
inference with small number of concurrent requests. vLLM's AWQ implementation have lower throughput than unquantized version.
To create a new 4-bit quantized model, you can leverage `AutoAWQ <https://github.com/casper-hansen/AutoAWQ>`_.
Quantizing reduces the model's precision from FP16 to INT4 which effectively reduces the file size by ~70%.
The main benefits are lower latency and memory usage.
You can quantize your own models by installing AutoAWQ or picking one of the `400+ models on Huggingface <https://huggingface.co/models?sort=trending&search=awq>`_.
.. code-block:: console
$ pip install autoawq
After installing AutoAWQ, you are ready to quantize a model. Here is an example of how to quantize `mistralai/Mistral-7B-Instruct-v0.2`:
.. code-block:: python
from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer
model_path = 'mistralai/Mistral-7B-Instruct-v0.2'
quant_path = 'mistral-instruct-v0.2-awq'
quant_config = { "zero_point": True, "q_group_size": 128, "w_bit": 4, "version": "GEMM" }
# Load model
model = AutoAWQForCausalLM.from_pretrained(
model_path, **{"low_cpu_mem_usage": True, "use_cache": False}
)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
# Quantize
model.quantize(tokenizer, quant_config=quant_config)
# Save quantized model
model.save_quantized(quant_path)
tokenizer.save_pretrained(quant_path)
print(f'Model is quantized and saved at "{quant_path}"')
To run an AWQ model with vLLM, you can use `TheBloke/Llama-2-7b-Chat-AWQ <https://huggingface.co/TheBloke/Llama-2-7b-Chat-AWQ>`_ with the following command:
.. code-block:: console
$ python examples/llm_engine_example.py --model TheBloke/Llama-2-7b-Chat-AWQ --quantization awq
AWQ models are also supported directly through the LLM entrypoint:
.. code-block:: python
from vllm import LLM, SamplingParams
# Sample prompts.
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Create an LLM.
llm = LLM(model="TheBloke/Llama-2-7b-Chat-AWQ", quantization="AWQ")
# Generate texts from the prompts. The output is a list of RequestOutput objects
# that contain the prompt, generated text, and other information.
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")

39
docs/source/quantization/bnb.md Normal file
View File

@ -0,0 +1,39 @@
(bits-and-bytes)=
# BitsAndBytes
vLLM now supports [BitsAndBytes](https://github.com/TimDettmers/bitsandbytes) for more efficient model inference.
BitsAndBytes quantizes models to reduce memory usage and enhance performance without significantly sacrificing accuracy.
Compared to other quantization methods, BitsAndBytes eliminates the need for calibrating the quantized model with input data.
Below are the steps to utilize BitsAndBytes with vLLM.
```console
$ pip install bitsandbytes>=0.45.0
```
vLLM reads the model's config file and supports both in-flight quantization and pre-quantized checkpoint.
You can find bitsandbytes quantized models on <https://huggingface.co/models?other=bitsandbytes>.
And usually, these repositories have a config.json file that includes a quantization_config section.
## Read quantized checkpoint.
```python
from vllm import LLM
import torch
# unsloth/tinyllama-bnb-4bit is a pre-quantized checkpoint.
model_id = "unsloth/tinyllama-bnb-4bit"
llm = LLM(model=model_id, dtype=torch.bfloat16, trust_remote_code=True, \
quantization="bitsandbytes", load_format="bitsandbytes")
```
## Inflight quantization: load as 4bit quantization
```python
from vllm import LLM
import torch
model_id = "huggyllama/llama-7b"
llm = LLM(model=model_id, dtype=torch.bfloat16, trust_remote_code=True, \
quantization="bitsandbytes", load_format="bitsandbytes")
```

43
docs/source/quantization/bnb.rst
View File

@ -1,43 +0,0 @@
.. _bits_and_bytes:
BitsAndBytes
==================
vLLM now supports `BitsAndBytes <https://github.com/TimDettmers/bitsandbytes>`_ for more efficient model inference.
BitsAndBytes quantizes models to reduce memory usage and enhance performance without significantly sacrificing accuracy.
Compared to other quantization methods, BitsAndBytes eliminates the need for calibrating the quantized model with input data.
Below are the steps to utilize BitsAndBytes with vLLM.
.. code-block:: console
$ pip install bitsandbytes>=0.45.0
vLLM reads the model's config file and supports both in-flight quantization and pre-quantized checkpoint.
You can find bitsandbytes quantized models on https://huggingface.co/models?other=bitsandbytes.
And usually, these repositories have a config.json file that includes a quantization_config section.
Read quantized checkpoint.
--------------------------
.. code-block:: python
from vllm import LLM
import torch
# unsloth/tinyllama-bnb-4bit is a pre-quantized checkpoint.
model_id = "unsloth/tinyllama-bnb-4bit"
llm = LLM(model=model_id, dtype=torch.bfloat16, trust_remote_code=True, \
quantization="bitsandbytes", load_format="bitsandbytes")
Inflight quantization: load as 4bit quantization
------------------------------------------------
.. code-block:: python
from vllm import LLM
import torch
model_id = "huggyllama/llama-7b"
llm = LLM(model=model_id, dtype=torch.bfloat16, trust_remote_code=True, \
quantization="bitsandbytes", load_format="bitsandbytes")

192
docs/source/quantization/fp8.md Normal file
View File

@ -0,0 +1,192 @@
(fp8)=
# FP8 W8A8
vLLM supports FP8 (8-bit floating point) weight and activation quantization using hardware acceleration on GPUs such as Nvidia H100 and AMD MI300x.
Currently, only Hopper and Ada Lovelace GPUs are officially supported for W8A8.
Ampere GPUs are supported for W8A16 (weight-only FP8) utilizing Marlin kernels.
Quantization of models with FP8 allows for a 2x reduction in model memory requirements and up to a 1.6x improvement in throughput with minimal impact on accuracy.
Please visit the HF collection of [quantized FP8 checkpoints of popular LLMs ready to use with vLLM](https://huggingface.co/collections/neuralmagic/fp8-llms-for-vllm-666742ed2b78b7ac8df13127).
The FP8 types typically supported in hardware have two distinct representations, each useful in different scenarios:
- **E4M3**: Consists of 1 sign bit, 4 exponent bits, and 3 bits of mantissa. It can store values up to +/-448 and `nan`.
- **E5M2**: Consists of 1 sign bit, 5 exponent bits, and 2 bits of mantissa. It can store values up to +/-57344, +/- `inf`, and `nan`. The tradeoff for the increased dynamic range is lower precision of the stored values.
```{note}
FP8 computation is supported on NVIDIA GPUs with compute capability > 8.9 (Ada Lovelace, Hopper).
FP8 models will run on compute capability > 8.0 (Ampere) as weight-only W8A16, utilizing FP8 Marlin.
```
## Quick Start with Online Dynamic Quantization
Dynamic quantization of an original precision BF16/FP16 model to FP8 can be achieved with vLLM without any calibration data required. You can enable the feature by specifying `--quantization="fp8"` in the command line or setting `quantization="fp8"` in the LLM constructor.
In this mode, all Linear modules (except for the final `lm_head`) have their weights quantized down to FP8_E4M3 precision with a per-tensor scale. Activations have their minimum and maximum values calculated during each forward pass to provide a dynamic per-tensor scale for high accuracy. As a result, latency improvements are limited in this mode.
```python
from vllm import LLM
model = LLM("facebook/opt-125m", quantization="fp8")
# INFO 06-10 17:55:42 model_runner.py:157] Loading model weights took 0.1550 GB
result = model.generate("Hello, my name is")
```
```{warning}
Currently, we load the model at original precision before quantizing down to 8-bits, so you need enough memory to load the whole model.
```
## Installation
To produce performant FP8 quantized models with vLLM, you'll need to install the [llm-compressor](https://github.com/vllm-project/llm-compressor/) library:
```console
$ pip install llmcompressor
```
## Quantization Process
The quantization process involves three main steps:
1. Loading the model
2. Applying quantization
3. Evaluating accuracy in vLLM
### 1. Loading the Model
Use `SparseAutoModelForCausalLM`, which wraps `AutoModelForCausalLM`, for saving and loading quantized models:
```python
from llmcompressor.transformers import SparseAutoModelForCausalLM
from transformers import AutoTokenizer
MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct"
model = SparseAutoModelForCausalLM.from_pretrained(
MODEL_ID, device_map="auto", torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
```
### 2. Applying Quantization
For FP8 quantization, we can recover accuracy with simple RTN quantization. We recommend targeting all `Linear` layers using the `FP8_DYNAMIC` scheme, which uses:
- Static, per-channel quantization on the weights
- Dynamic, per-token quantization on the activations
Since simple RTN does not require data for weight quantization and the activations are quantized dynamically, we do not need any calibration data for this quantization flow.
```python
from llmcompressor.transformers import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
# Configure the simple PTQ quantization
recipe = QuantizationModifier(
targets="Linear", scheme="FP8_DYNAMIC", ignore=["lm_head"])
# Apply the quantization algorithm.
oneshot(model=model, recipe=recipe)
# Save the model.
SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-Dynamic"
model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)
```
### 3. Evaluating Accuracy
Install `vllm` and `lm-evaluation-harness`:
```console
$ pip install vllm lm-eval==0.4.4
```
Load and run the model in `vllm`:
```python
from vllm import LLM
model = LLM("./Meta-Llama-3-8B-Instruct-FP8-Dynamic")
model.generate("Hello my name is")
```
Evaluate accuracy with `lm_eval` (for example on 250 samples of `gsm8k`):
```{note}
Quantized models can be sensitive to the presence of the `bos` token. `lm_eval` does not add a `bos` token by default, so make sure to include the `add_bos_token=True` argument when running your evaluations.
```
```console
$ MODEL=$PWD/Meta-Llama-3-8B-Instruct-FP8-Dynamic
$ lm_eval \
--model vllm \
--model_args pretrained=$MODEL,add_bos_token=True \
--tasks gsm8k --num_fewshot 5 --batch_size auto --limit 250
```
Here's an example of the resulting scores:
```text
|Tasks|Version| Filter |n-shot| Metric | |Value| |Stderr|
|-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k| 3|flexible-extract| 5|exact_match|↑ |0.768|± |0.0268|
| | |strict-match | 5|exact_match|↑ |0.768|± |0.0268|
```
## Troubleshooting and Support
If you encounter any issues or have feature requests, please open an issue on the `vllm-project/llm-compressor` GitHub repository.
## Deprecated Flow
```{note}
The following information is preserved for reference and search purposes.
The quantization method described below is deprecated in favor of the `llmcompressor` method described above.
```
For static per-tensor offline quantization to FP8, please install the [AutoFP8 library](https://github.com/neuralmagic/autofp8).
```bash
git clone https://github.com/neuralmagic/AutoFP8.git
pip install -e AutoFP8
```
This package introduces the `AutoFP8ForCausalLM` and `BaseQuantizeConfig` objects for managing how your model will be compressed.
## Offline Quantization with Static Activation Scaling Factors
You can use AutoFP8 with calibration data to produce per-tensor static scales for both the weights and activations by enabling the `activation_scheme="static"` argument.
```python
from datasets import load_dataset
from transformers import AutoTokenizer
from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
pretrained_model_dir = "meta-llama/Meta-Llama-3-8B-Instruct"
quantized_model_dir = "Meta-Llama-3-8B-Instruct-FP8"
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True)
tokenizer.pad_token = tokenizer.eos_token
# Load and tokenize 512 dataset samples for calibration of activation scales
ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(range(512))
examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
# Define quantization config with static activation scales
quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="static")
# Load the model, quantize, and save checkpoint
model = AutoFP8ForCausalLM.from_pretrained(pretrained_model_dir, quantize_config)
model.quantize(examples)
model.save_quantized(quantized_model_dir)
```
Your model checkpoint with quantized weights and activations should be available at `Meta-Llama-3-8B-Instruct-FP8/`.
Finally, you can load the quantized model checkpoint directly in vLLM.
```python
from vllm import LLM
model = LLM(model="Meta-Llama-3-8B-Instruct-FP8/")
# INFO 06-10 21:15:41 model_runner.py:159] Loading model weights took 8.4596 GB
result = model.generate("Hello, my name is")
```

204
docs/source/quantization/fp8.rst
View File

@ -1,204 +0,0 @@
.. _fp8:
FP8 W8A8
==================
vLLM supports FP8 (8-bit floating point) weight and activation quantization using hardware acceleration on GPUs such as Nvidia H100 and AMD MI300x.
Currently, only Hopper and Ada Lovelace GPUs are officially supported for W8A8.
Ampere GPUs are supported for W8A16 (weight-only FP8) utilizing Marlin kernels.
Quantization of models with FP8 allows for a 2x reduction in model memory requirements and up to a 1.6x improvement in throughput with minimal impact on accuracy.
Please visit the HF collection of `quantized FP8 checkpoints of popular LLMs ready to use with vLLM <https://huggingface.co/collections/neuralmagic/fp8-llms-for-vllm-666742ed2b78b7ac8df13127>`_.
The FP8 types typically supported in hardware have two distinct representations, each useful in different scenarios:
- **E4M3**: Consists of 1 sign bit, 4 exponent bits, and 3 bits of mantissa. It can store values up to +/-448 and ``nan``.
- **E5M2**: Consists of 1 sign bit, 5 exponent bits, and 2 bits of mantissa. It can store values up to +/-57344, +/- ``inf``, and ``nan``. The tradeoff for the increased dynamic range is lower precision of the stored values.
.. note::
FP8 computation is supported on NVIDIA GPUs with compute capability > 8.9 (Ada Lovelace, Hopper).
FP8 models will run on compute capability > 8.0 (Ampere) as weight-only W8A16, utilizing FP8 Marlin.
Quick Start with Online Dynamic Quantization
--------------------------------------------
Dynamic quantization of an original precision BF16/FP16 model to FP8 can be achieved with vLLM without any calibration data required. You can enable the feature by specifying ``--quantization="fp8"`` in the command line or setting ``quantization="fp8"`` in the LLM constructor.
In this mode, all Linear modules (except for the final ``lm_head``) have their weights quantized down to FP8_E4M3 precision with a per-tensor scale. Activations have their minimum and maximum values calculated during each forward pass to provide a dynamic per-tensor scale for high accuracy. As a result, latency improvements are limited in this mode.
.. code-block:: python
from vllm import LLM
model = LLM("facebook/opt-125m", quantization="fp8")
# INFO 06-10 17:55:42 model_runner.py:157] Loading model weights took 0.1550 GB
result = model.generate("Hello, my name is")
.. warning::
Currently, we load the model at original precision before quantizing down to 8-bits, so you need enough memory to load the whole model.
Installation
------------
To produce performant FP8 quantized models with vLLM, you'll need to install the `llm-compressor <https://github.com/vllm-project/llm-compressor/>`_ library:
.. code-block:: console
$ pip install llmcompressor
Quantization Process
--------------------
The quantization process involves three main steps:
1. Loading the model
2. Applying quantization
3. Evaluating accuracy in vLLM
1. Loading the Model
^^^^^^^^^^^^^^^^^^^^
Use ``SparseAutoModelForCausalLM``, which wraps ``AutoModelForCausalLM``, for saving and loading quantized models:
.. code-block:: python
from llmcompressor.transformers import SparseAutoModelForCausalLM
from transformers import AutoTokenizer
MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct"
model = SparseAutoModelForCausalLM.from_pretrained(
MODEL_ID, device_map="auto", torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
2. Applying Quantization
^^^^^^^^^^^^^^^^^^^^^^^^
For FP8 quantization, we can recover accuracy with simple RTN quantization. We recommend targeting all ``Linear`` layers using the ``FP8_DYNAMIC`` scheme, which uses:
- Static, per-channel quantization on the weights
- Dynamic, per-token quantization on the activations
Since simple RTN does not require data for weight quantization and the activations are quantized dynamically, we do not need any calibration data for this quantization flow.
.. code-block:: python
from llmcompressor.transformers import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
# Configure the simple PTQ quantization
recipe = QuantizationModifier(
targets="Linear", scheme="FP8_DYNAMIC", ignore=["lm_head"])
# Apply the quantization algorithm.
oneshot(model=model, recipe=recipe)
# Save the model.
SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8-Dynamic"
model.save_pretrained(SAVE_DIR)
tokenizer.save_pretrained(SAVE_DIR)
3. Evaluating Accuracy
^^^^^^^^^^^^^^^^^^^^^^
Install ``vllm`` and ``lm-evaluation-harness``:
.. code-block:: console
$ pip install vllm lm-eval==0.4.4
Load and run the model in ``vllm``:
.. code-block:: python
from vllm import LLM
model = LLM("./Meta-Llama-3-8B-Instruct-FP8-Dynamic")
model.generate("Hello my name is")
Evaluate accuracy with ``lm_eval`` (for example on 250 samples of ``gsm8k``):
.. note::
Quantized models can be sensitive to the presence of the ``bos`` token. ``lm_eval`` does not add a ``bos`` token by default, so make sure to include the ``add_bos_token=True`` argument when running your evaluations.
.. code-block:: console
$ MODEL=$PWD/Meta-Llama-3-8B-Instruct-FP8-Dynamic
$ lm_eval \
--model vllm \
--model_args pretrained=$MODEL,add_bos_token=True \
--tasks gsm8k --num_fewshot 5 --batch_size auto --limit 250
Here's an example of the resulting scores:
.. code-block:: text
|Tasks|Version| Filter |n-shot| Metric | |Value| |Stderr|
|-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k| 3|flexible-extract| 5|exact_match|↑ |0.768|± |0.0268|
| | |strict-match | 5|exact_match|↑ |0.768|± |0.0268|
Troubleshooting and Support
---------------------------
If you encounter any issues or have feature requests, please open an issue on the ``vllm-project/llm-compressor`` GitHub repository.
Deprecated Flow
------------------
.. note::
The following information is preserved for reference and search purposes.
The quantization method described below is deprecated in favor of the ``llmcompressor`` method described above.
For static per-tensor offline quantization to FP8, please install the `AutoFP8 library <https://github.com/neuralmagic/autofp8>`_.
.. code-block:: bash
git clone https://github.com/neuralmagic/AutoFP8.git
pip install -e AutoFP8
This package introduces the ``AutoFP8ForCausalLM`` and ``BaseQuantizeConfig`` objects for managing how your model will be compressed.
Offline Quantization with Static Activation Scaling Factors
-----------------------------------------------------------
You can use AutoFP8 with calibration data to produce per-tensor static scales for both the weights and activations by enabling the ``activation_scheme="static"`` argument.
.. code-block:: python
from datasets import load_dataset
from transformers import AutoTokenizer
from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
pretrained_model_dir = "meta-llama/Meta-Llama-3-8B-Instruct"
quantized_model_dir = "Meta-Llama-3-8B-Instruct-FP8"
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True)
tokenizer.pad_token = tokenizer.eos_token
# Load and tokenize 512 dataset samples for calibration of activation scales
ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(range(512))
examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
# Define quantization config with static activation scales
quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="static")
# Load the model, quantize, and save checkpoint
model = AutoFP8ForCausalLM.from_pretrained(pretrained_model_dir, quantize_config)
model.quantize(examples)
model.save_quantized(quantized_model_dir)
Your model checkpoint with quantized weights and activations should be available at ``Meta-Llama-3-8B-Instruct-FP8/``.
Finally, you can load the quantized model checkpoint directly in vLLM.
.. code-block:: python
from vllm import LLM
model = LLM(model="Meta-Llama-3-8B-Instruct-FP8/")
# INFO 06-10 21:15:41 model_runner.py:159] Loading model weights took 8.4596 GB
result = model.generate("Hello, my name is")

44
docs/source/quantization/fp8_e4m3_kvcache.md Normal file
View File

@ -0,0 +1,44 @@
(fp8-e4m3-kvcache)=
# FP8 E4M3 KV Cache
Quantizing the KV cache to FP8 reduces its memory footprint. This increases the number of tokens that can be stored in the cache,
improving throughput. OCP (Open Compute Project www.opencompute.org) specifies two common 8-bit floating point data formats: E5M2
(5 exponent bits and 2 mantissa bits) and E4M3FN (4 exponent bits and 3 mantissa bits), often shortened as E4M3. One benefit of
the E4M3 format over E5M2 is that floating point numbers are represented in higher precision. However, the small dynamic range of
FP8 E4M3 (±240.0 can be represented) typically necessitates the use of a higher-precision (typically FP32) scaling factor alongside
each quantized tensor. For now, only per-tensor (scalar) scaling factors are supported. Development is ongoing to support scaling
factors of a finer granularity (e.g. per-channel).
These scaling factors can be specified by passing an optional quantization param JSON to the LLM engine at load time. If
this JSON is not specified, scaling factors default to 1.0. These scaling factors are typically obtained when running an
unquantized model through a quantizer tool (e.g. AMD quantizer or NVIDIA AMMO).
To install AMMO (AlgorithMic Model Optimization):
```console
$ pip install --no-cache-dir --extra-index-url https://pypi.nvidia.com nvidia-ammo
```
Studies have shown that FP8 E4M3 quantization typically only minimally degrades inference accuracy. The most recent silicon
offerings e.g. AMD MI300, NVIDIA Hopper or later support native hardware conversion to and from fp32, fp16, bf16, etc.
Thus, LLM inference is greatly accelerated with minimal accuracy loss.
Here is an example of how to enable this feature:
```python
# two float8_e4m3fn kv cache scaling factor files are provided under tests/fp8_kv, please refer to
# https://github.com/vllm-project/vllm/blob/main/examples/fp8/README.md to generate kv_cache_scales.json of your own.
from vllm import LLM, SamplingParams
sampling_params = SamplingParams(temperature=1.3, top_p=0.8)
llm = LLM(model="meta-llama/Llama-2-7b-chat-hf",
kv_cache_dtype="fp8",
quantization_param_path="./tests/fp8_kv/llama2-7b-fp8-kv/kv_cache_scales.json")
prompt = "London is the capital of"
out = llm.generate(prompt, sampling_params)[0].outputs[0].text
print(out)
# output w/ scaling factors: England, the United Kingdom, and one of the world's leading financial,
# output w/o scaling factors: England, located in the southeastern part of the country. It is known
```

47
docs/source/quantization/fp8_e4m3_kvcache.rst
View File

@ -1,47 +0,0 @@
.. _fp8_e4m3_kvcache:
FP8 E4M3 KV Cache
==================
Quantizing the KV cache to FP8 reduces its memory footprint. This increases the number of tokens that can be stored in the cache,
improving throughput. OCP (Open Compute Project www.opencompute.org) specifies two common 8-bit floating point data formats: E5M2
(5 exponent bits and 2 mantissa bits) and E4M3FN (4 exponent bits and 3 mantissa bits), often shortened as E4M3. One benefit of
the E4M3 format over E5M2 is that floating point numbers are represented in higher precision. However, the small dynamic range of
FP8 E4M3 (±240.0 can be represented) typically necessitates the use of a higher-precision (typically FP32) scaling factor alongside
each quantized tensor. For now, only per-tensor (scalar) scaling factors are supported. Development is ongoing to support scaling
factors of a finer granularity (e.g. per-channel).
These scaling factors can be specified by passing an optional quantization param JSON to the LLM engine at load time. If
this JSON is not specified, scaling factors default to 1.0. These scaling factors are typically obtained when running an
unquantized model through a quantizer tool (e.g. AMD quantizer or NVIDIA AMMO).
To install AMMO (AlgorithMic Model Optimization):
.. code-block:: console
$ pip install --no-cache-dir --extra-index-url https://pypi.nvidia.com nvidia-ammo
Studies have shown that FP8 E4M3 quantization typically only minimally degrades inference accuracy. The most recent silicon
offerings e.g. AMD MI300, NVIDIA Hopper or later support native hardware conversion to and from fp32, fp16, bf16, etc.
Thus, LLM inference is greatly accelerated with minimal accuracy loss.
Here is an example of how to enable this feature:
.. code-block:: python
# two float8_e4m3fn kv cache scaling factor files are provided under tests/fp8_kv, please refer to
# https://github.com/vllm-project/vllm/blob/main/examples/fp8/README.md to generate kv_cache_scales.json of your own.
from vllm import LLM, SamplingParams
sampling_params = SamplingParams(temperature=1.3, top_p=0.8)
llm = LLM(model="meta-llama/Llama-2-7b-chat-hf",
kv_cache_dtype="fp8",
quantization_param_path="./tests/fp8_kv/llama2-7b-fp8-kv/kv_cache_scales.json")
prompt = "London is the capital of"
out = llm.generate(prompt, sampling_params)[0].outputs[0].text
print(out)
# output w/ scaling factors: England, the United Kingdom, and one of the world's leading financial,
# output w/o scaling factors: England, located in the southeastern part of the country. It is known

31
docs/source/quantization/fp8_e5m2_kvcache.md Normal file
View File

@ -0,0 +1,31 @@
(fp8-kv-cache)=
# FP8 E5M2 KV Cache
The int8/int4 quantization scheme requires additional scale GPU memory storage, which reduces the expected GPU memory benefits.
The FP8 data format retains 2~3 mantissa bits and can convert float/fp16/bfloat16 and fp8 to each other.
Here is an example of how to enable this feature:
```python
from vllm import LLM, SamplingParams
# Sample prompts.
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Create an LLM.
llm = LLM(model="facebook/opt-125m", kv_cache_dtype="fp8")
# Generate texts from the prompts. The output is a list of RequestOutput objects
# that contain the prompt, generated text, and other information.
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```

34
docs/source/quantization/fp8_e5m2_kvcache.rst
View File

@ -1,34 +0,0 @@
.. _fp8_kv_cache:
FP8 E5M2 KV Cache
==================
The int8/int4 quantization scheme requires additional scale GPU memory storage, which reduces the expected GPU memory benefits.
The FP8 data format retains 2~3 mantissa bits and can convert float/fp16/bfloat16 and fp8 to each other.
Here is an example of how to enable this feature:
.. code-block:: python
from vllm import LLM, SamplingParams
# Sample prompts.
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Create an LLM.
llm = LLM(model="facebook/opt-125m", kv_cache_dtype="fp8")
# Generate texts from the prompts. The output is a list of RequestOutput objects
# that contain the prompt, generated text, and other information.
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")

72
docs/source/quantization/gguf.md Normal file
View File

@ -0,0 +1,72 @@
(gguf)=
# GGUF
```{warning}
Please note that GGUF support in vLLM is highly experimental and under-optimized at the moment, it might be incompatible with other features. Currently, you can use GGUF as a way to reduce memory footprint. If you encounter any issues, please report them to the vLLM team.
```
```{warning}
Currently, vllm only supports loading single-file GGUF models. If you have a multi-files GGUF model, you can use [gguf-split](https://github.com/ggerganov/llama.cpp/pull/6135) tool to merge them to a single-file model.
```
To run a GGUF model with vLLM, you can download and use the local GGUF model from [TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF](https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF) with the following command:
```console
$ wget https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf
$ # We recommend using the tokenizer from base model to avoid long-time and buggy tokenizer conversion.
$ vllm serve ./tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf --tokenizer TinyLlama/TinyLlama-1.1B-Chat-v1.0
```
You can also add `--tensor-parallel-size 2` to enable tensor parallelism inference with 2 GPUs:
```console
$ # We recommend using the tokenizer from base model to avoid long-time and buggy tokenizer conversion.
$ vllm serve ./tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf --tokenizer TinyLlama/TinyLlama-1.1B-Chat-v1.0 --tensor-parallel-size 2
```
```{warning}
We recommend using the tokenizer from base model instead of GGUF model. Because the tokenizer conversion from GGUF is time-consuming and unstable, especially for some models with large vocab size.
```
You can also use the GGUF model directly through the LLM entrypoint:
```python
from vllm import LLM, SamplingParams
# In this script, we demonstrate how to pass input to the chat method:
conversation = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": "Hello"
},
{
"role": "assistant",
"content": "Hello! How can I assist you today?"
},
{
"role": "user",
"content": "Write an essay about the importance of higher education.",
},
]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Create an LLM.
llm = LLM(model="./tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
tokenizer="TinyLlama/TinyLlama-1.1B-Chat-v1.0")
# Generate texts from the prompts. The output is a list of RequestOutput objects
# that contain the prompt, generated text, and other information.
outputs = llm.chat(conversation, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```

73
docs/source/quantization/gguf.rst
View File

@ -1,73 +0,0 @@
.. _gguf:
GGUF
==================
.. warning::
Please note that GGUF support in vLLM is highly experimental and under-optimized at the moment, it might be incompatible with other features. Currently, you can use GGUF as a way to reduce memory footprint. If you encounter any issues, please report them to the vLLM team.
.. warning::
Currently, vllm only supports loading single-file GGUF models. If you have a multi-files GGUF model, you can use `gguf-split <https://github.com/ggerganov/llama.cpp/pull/6135>`_ tool to merge them to a single-file model.
To run a GGUF model with vLLM, you can download and use the local GGUF model from `TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF <https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF>`_ with the following command:
.. code-block:: console
$ wget https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf
$ # We recommend using the tokenizer from base model to avoid long-time and buggy tokenizer conversion.
$ vllm serve ./tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf --tokenizer TinyLlama/TinyLlama-1.1B-Chat-v1.0
You can also add ``--tensor-parallel-size 2`` to enable tensor parallelism inference with 2 GPUs:
.. code-block:: console
$ # We recommend using the tokenizer from base model to avoid long-time and buggy tokenizer conversion.
$ vllm serve ./tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf --tokenizer TinyLlama/TinyLlama-1.1B-Chat-v1.0 --tensor-parallel-size 2
.. warning::
We recommend using the tokenizer from base model instead of GGUF model. Because the tokenizer conversion from GGUF is time-consuming and unstable, especially for some models with large vocab size.
You can also use the GGUF model directly through the LLM entrypoint:
.. code-block:: python
from vllm import LLM, SamplingParams
# In this script, we demonstrate how to pass input to the chat method:
conversation = [
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": "Hello"
},
{
"role": "assistant",
"content": "Hello! How can I assist you today?"
},
{
"role": "user",
"content": "Write an essay about the importance of higher education.",
},
]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
# Create an LLM.
llm = LLM(model="./tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf",
tokenizer="TinyLlama/TinyLlama-1.1B-Chat-v1.0")
# Generate texts from the prompts. The output is a list of RequestOutput objects
# that contain the prompt, generated text, and other information.
outputs = llm.chat(conversation, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")

136
docs/source/quantization/int8.md Normal file
View File

@ -0,0 +1,136 @@
(int8)=
# INT8 W8A8
vLLM supports quantizing weights and activations to INT8 for memory savings and inference acceleration.
This quantization method is particularly useful for reducing model size while maintaining good performance.
Please visit the HF collection of [quantized INT8 checkpoints of popular LLMs ready to use with vLLM](https://huggingface.co/collections/neuralmagic/int8-llms-for-vllm-668ec32c049dca0369816415).
```{note}
INT8 computation is supported on NVIDIA GPUs with compute capability > 7.5 (Turing, Ampere, Ada Lovelace, Hopper).
```
## Prerequisites
To use INT8 quantization with vLLM, you'll need to install the [llm-compressor](https://github.com/vllm-project/llm-compressor/) library:
```console
$ pip install llmcompressor
```
## Quantization Process
The quantization process involves four main steps:
1. Loading the model
2. Preparing calibration data
3. Applying quantization
4. Evaluating accuracy in vLLM
### 1. Loading the Model
Use `SparseAutoModelForCausalLM`, which wraps `AutoModelForCausalLM`, for saving and loading quantized models:
```python
from llmcompressor.transformers import SparseAutoModelForCausalLM
from transformers import AutoTokenizer
MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct"
model = SparseAutoModelForCausalLM.from_pretrained(
MODEL_ID, device_map="auto", torch_dtype="auto",
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
```
### 2. Preparing Calibration Data
When quantizing activations to INT8, you need sample data to estimate the activation scales.
It's best to use calibration data that closely matches your deployment data.
For a general-purpose instruction-tuned model, you can use a dataset like `ultrachat`:
```python
from datasets import load_dataset
NUM_CALIBRATION_SAMPLES = 512
MAX_SEQUENCE_LENGTH = 2048
# Load and preprocess the dataset
ds = load_dataset("HuggingFaceH4/ultrachat_200k", split="train_sft")
ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))
def preprocess(example):
return {"text": tokenizer.apply_chat_template(example["messages"], tokenize=False)}
ds = ds.map(preprocess)
def tokenize(sample):
return tokenizer(sample["text"], padding=False, max_length=MAX_SEQUENCE_LENGTH, truncation=True, add_special_tokens=False)
ds = ds.map(tokenize, remove_columns=ds.column_names)
```
### 3. Applying Quantization
Now, apply the quantization algorithms:
```python
from llmcompressor.transformers import oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
# Configure the quantization algorithms
recipe = [
SmoothQuantModifier(smoothing_strength=0.8),
GPTQModifier(targets="Linear", scheme="W8A8", ignore=["lm_head"]),
]
# Apply quantization
oneshot(
model=model,
dataset=ds,
recipe=recipe,
max_seq_length=MAX_SEQUENCE_LENGTH,
num_calibration_samples=NUM_CALIBRATION_SAMPLES,
)
# Save the compressed model
SAVE_DIR = MODEL_ID.split("/")[1] + "-W8A8-Dynamic-Per-Token"
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)
```
This process creates a W8A8 model with weights and activations quantized to 8-bit integers.
### 4. Evaluating Accuracy
After quantization, you can load and run the model in vLLM:
```python
from vllm import LLM
model = LLM("./Meta-Llama-3-8B-Instruct-W8A8-Dynamic-Per-Token")
```
To evaluate accuracy, you can use `lm_eval`:
```console
$ lm_eval --model vllm \
--model_args pretrained="./Meta-Llama-3-8B-Instruct-W8A8-Dynamic-Per-Token",add_bos_token=true \
--tasks gsm8k \
--num_fewshot 5 \
--limit 250 \
--batch_size 'auto'
```
```{note}
Quantized models can be sensitive to the presence of the `bos` token. Make sure to include the `add_bos_token=True` argument when running evaluations.
```
## Best Practices
- Start with 512 samples for calibration data (increase if accuracy drops)
- Use a sequence length of 2048 as a starting point
- Employ the chat template or instruction template that the model was trained with
- If you've fine-tuned a model, consider using a sample of your training data for calibration
## Troubleshooting and Support
If you encounter any issues or have feature requests, please open an issue on the `vllm-project/llm-compressor` GitHub repository.

145
docs/source/quantization/int8.rst
View File

@ -1,145 +0,0 @@
.. _int8:
INT8 W8A8
==================
vLLM supports quantizing weights and activations to INT8 for memory savings and inference acceleration.
This quantization method is particularly useful for reducing model size while maintaining good performance.
Please visit the HF collection of `quantized INT8 checkpoints of popular LLMs ready to use with vLLM <https://huggingface.co/collections/neuralmagic/int8-llms-for-vllm-668ec32c049dca0369816415>`_.
.. note::
INT8 computation is supported on NVIDIA GPUs with compute capability > 7.5 (Turing, Ampere, Ada Lovelace, Hopper).
Prerequisites
-------------
To use INT8 quantization with vLLM, you'll need to install the `llm-compressor <https://github.com/vllm-project/llm-compressor/>`_ library:
.. code-block:: console
$ pip install llmcompressor
Quantization Process
--------------------
The quantization process involves four main steps:
1. Loading the model
2. Preparing calibration data
3. Applying quantization
4. Evaluating accuracy in vLLM
1. Loading the Model
^^^^^^^^^^^^^^^^^^^^
Use ``SparseAutoModelForCausalLM``, which wraps ``AutoModelForCausalLM``, for saving and loading quantized models:
.. code-block:: python
from llmcompressor.transformers import SparseAutoModelForCausalLM
from transformers import AutoTokenizer
MODEL_ID = "meta-llama/Meta-Llama-3-8B-Instruct"
model = SparseAutoModelForCausalLM.from_pretrained(
MODEL_ID, device_map="auto", torch_dtype="auto",
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
2. Preparing Calibration Data
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
When quantizing activations to INT8, you need sample data to estimate the activation scales.
It's best to use calibration data that closely matches your deployment data.
For a general-purpose instruction-tuned model, you can use a dataset like ``ultrachat``:
.. code-block:: python
from datasets import load_dataset
NUM_CALIBRATION_SAMPLES = 512
MAX_SEQUENCE_LENGTH = 2048
# Load and preprocess the dataset
ds = load_dataset("HuggingFaceH4/ultrachat_200k", split="train_sft")
ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))
def preprocess(example):
return {"text": tokenizer.apply_chat_template(example["messages"], tokenize=False)}
ds = ds.map(preprocess)
def tokenize(sample):
return tokenizer(sample["text"], padding=False, max_length=MAX_SEQUENCE_LENGTH, truncation=True, add_special_tokens=False)
ds = ds.map(tokenize, remove_columns=ds.column_names)
3. Applying Quantization
^^^^^^^^^^^^^^^^^^^^^^^^
Now, apply the quantization algorithms:
.. code-block:: python
from llmcompressor.transformers import oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
# Configure the quantization algorithms
recipe = [
SmoothQuantModifier(smoothing_strength=0.8),
GPTQModifier(targets="Linear", scheme="W8A8", ignore=["lm_head"]),
]
# Apply quantization
oneshot(
model=model,
dataset=ds,
recipe=recipe,
max_seq_length=MAX_SEQUENCE_LENGTH,
num_calibration_samples=NUM_CALIBRATION_SAMPLES,
)
# Save the compressed model
SAVE_DIR = MODEL_ID.split("/")[1] + "-W8A8-Dynamic-Per-Token"
model.save_pretrained(SAVE_DIR, save_compressed=True)
tokenizer.save_pretrained(SAVE_DIR)
This process creates a W8A8 model with weights and activations quantized to 8-bit integers.
4. Evaluating Accuracy
^^^^^^^^^^^^^^^^^^^^^^
After quantization, you can load and run the model in vLLM:
.. code-block:: python
from vllm import LLM
model = LLM("./Meta-Llama-3-8B-Instruct-W8A8-Dynamic-Per-Token")
To evaluate accuracy, you can use ``lm_eval``:
.. code-block:: console
$ lm_eval --model vllm \
--model_args pretrained="./Meta-Llama-3-8B-Instruct-W8A8-Dynamic-Per-Token",add_bos_token=true \
--tasks gsm8k \
--num_fewshot 5 \
--limit 250 \
--batch_size 'auto'
.. note::
Quantized models can be sensitive to the presence of the ``bos`` token. Make sure to include the ``add_bos_token=True`` argument when running evaluations.
Best Practices
--------------
- Start with 512 samples for calibration data (increase if accuracy drops)
- Use a sequence length of 2048 as a starting point
- Employ the chat template or instruction template that the model was trained with
- If you've fine-tuned a model, consider using a sample of your training data for calibration
Troubleshooting and Support
---------------------------
If you encounter any issues or have feature requests, please open an issue on the ``vllm-project/llm-compressor`` GitHub repository.

12
docs/source/quantization/supported_hardware.rst → docs/source/quantization/supported_hardware.md
View File

@ -1,10 +1,10 @@
.. _supported_hardware_for_quantization:
(supported-hardware-for-quantization)=
Supported Hardware for Quantization Kernels
===========================================
# Supported Hardware for Quantization Kernels
The table below shows the compatibility of various quantization implementations with different hardware platforms in vLLM:
```{eval-rst}
.. list-table::
:header-rows: 1
:widths: 20 8 8 8 8 8 8 8 8 8 8
@ -119,9 +119,9 @@ The table below shows the compatibility of various quantization implementations
- ✗
- ✗
- ✗
```
Notes:
^^^^^^
## Notes:
- Volta refers to SM 7.0, Turing to SM 7.5, Ampere to SM 8.0/8.6, Ada to SM 8.9, and Hopper to SM 9.0.
- "✅︎" indicates that the quantization method is supported on the specified hardware.
@ -129,4 +129,4 @@ Notes:
Please note that this compatibility chart may be subject to change as vLLM continues to evolve and expand its support for different hardware platforms and quantization methods.
For the most up-to-date information on hardware support and quantization methods, please check the `quantization directory <https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/quantization>`_ or consult with the vLLM development team.
For the most up-to-date information on hardware support and quantization methods, please check the [quantization directory](https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/quantization) or consult with the vLLM development team.

7
docs/source/serving/deploying_with_bentoml.md Normal file
View File

@ -0,0 +1,7 @@
(deploying-with-bentoml)=
# Deploying with BentoML
[BentoML](https://github.com/bentoml/BentoML) allows you to deploy a large language model (LLM) server with vLLM as the backend, which exposes OpenAI-compatible endpoints. You can serve the model locally or containerize it as an OCI-complicant image and deploy it on Kubernetes.
For details, see the tutorial [vLLM inference in the BentoML documentation](https://docs.bentoml.com/en/latest/use-cases/large-language-models/vllm.html).

8
docs/source/serving/deploying_with_bentoml.rst
View File

@ -1,8 +0,0 @@
.. _deploying_with_bentoml:
Deploying with BentoML
======================
`BentoML <https://github.com/bentoml/BentoML>`_ allows you to deploy a large language model (LLM) server with vLLM as the backend, which exposes OpenAI-compatible endpoints. You can serve the model locally or containerize it as an OCI-complicant image and deploy it on Kubernetes.
For details, see the tutorial `vLLM inference in the BentoML documentation <https://docs.bentoml.com/en/latest/use-cases/large-language-models/vllm.html>`_.

109
docs/source/serving/deploying_with_cerebrium.md Normal file
View File

@ -0,0 +1,109 @@
(deploying-with-cerebrium)=
# Deploying with Cerebrium
```{raw} html
<p align="center">
<img src="https://i.ibb.co/hHcScTT/Screenshot-2024-06-13-at-10-14-54.png" alt="vLLM_plus_cerebrium"/>
</p>
```
vLLM can be run on a cloud based GPU machine with [Cerebrium](https://www.cerebrium.ai/), a serverless AI infrastructure platform that makes it easier for companies to build and deploy AI based applications.
To install the Cerebrium client, run:
```console
$ pip install cerebrium
$ cerebrium login
```
Next, create your Cerebrium project, run:
```console
$ cerebrium init vllm-project
```
Next, to install the required packages, add the following to your cerebrium.toml:
```toml
[cerebrium.deployment]
docker_base_image_url = "nvidia/cuda:12.1.1-runtime-ubuntu22.04"
[cerebrium.dependencies.pip]
vllm = "latest"
```
Next, let us add our code to handle inference for the LLM of your choice(`mistralai/Mistral-7B-Instruct-v0.1` for this example), add the following code to your main.py\`:
```python
from vllm import LLM, SamplingParams
llm = LLM(model="mistralai/Mistral-7B-Instruct-v0.1")
def run(prompts: list[str], temperature: float = 0.8, top_p: float = 0.95):
sampling_params = SamplingParams(temperature=temperature, top_p=top_p)
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
results = []
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
results.append({"prompt": prompt, "generated_text": generated_text})
return {"results": results}
```
Then, run the following code to deploy it to the cloud
```console
$ cerebrium deploy
```
If successful, you should be returned a CURL command that you can call inference against. Just remember to end the url with the function name you are calling (in our case /run)
```python
curl -X POST https://api.cortex.cerebrium.ai/v4/p-xxxxxx/vllm/run \
-H 'Content-Type: application/json' \
-H 'Authorization: <JWT TOKEN>' \
--data '{
"prompts": [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is"
]
}'
```
You should get a response like:
```python
{
"run_id": "52911756-3066-9ae8-bcc9-d9129d1bd262",
"result": {
"result": [
{
"prompt": "Hello, my name is",
"generated_text": " Sarah, and I'm a teacher. I teach elementary school students. One of"
},
{
"prompt": "The president of the United States is",
"generated_text": " elected every four years. This is a democratic system.\n\n5. What"
},
{
"prompt": "The capital of France is",
"generated_text": " Paris.\n"
},
{
"prompt": "The future of AI is",
"generated_text": " bright, but it's important to approach it with a balanced and nuanced perspective."
}
]
},
"run_time_ms": 152.53663063049316
}
```
You now have an autoscaling endpoint where you only pay for the compute you use!

112
docs/source/serving/deploying_with_cerebrium.rst
View File

@ -1,112 +0,0 @@
.. _deploying_with_cerebrium:
Deploying with Cerebrium
============================
.. raw:: html
<p align="center">
<img src="https://i.ibb.co/hHcScTT/Screenshot-2024-06-13-at-10-14-54.png" alt="vLLM_plus_cerebrium"/>
</p>
vLLM can be run on a cloud based GPU machine with `Cerebrium <https://www.cerebrium.ai/>`__, a serverless AI infrastructure platform that makes it easier for companies to build and deploy AI based applications.
To install the Cerebrium client, run:
.. code-block:: console
$ pip install cerebrium
$ cerebrium login
Next, create your Cerebrium project, run:
.. code-block:: console
$ cerebrium init vllm-project
Next, to install the required packages, add the following to your cerebrium.toml:
.. code-block:: toml
[cerebrium.deployment]
docker_base_image_url = "nvidia/cuda:12.1.1-runtime-ubuntu22.04"
[cerebrium.dependencies.pip]
vllm = "latest"
Next, let us add our code to handle inference for the LLM of your choice(`mistralai/Mistral-7B-Instruct-v0.1` for this example), add the following code to your main.py`:
.. code-block:: python
from vllm import LLM, SamplingParams
llm = LLM(model="mistralai/Mistral-7B-Instruct-v0.1")
def run(prompts: list[str], temperature: float = 0.8, top_p: float = 0.95):
sampling_params = SamplingParams(temperature=temperature, top_p=top_p)
outputs = llm.generate(prompts, sampling_params)
# Print the outputs.
results = []
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
results.append({"prompt": prompt, "generated_text": generated_text})
return {"results": results}
Then, run the following code to deploy it to the cloud
.. code-block:: console
$ cerebrium deploy
If successful, you should be returned a CURL command that you can call inference against. Just remember to end the url with the function name you are calling (in our case /run)
.. code-block:: python
curl -X POST https://api.cortex.cerebrium.ai/v4/p-xxxxxx/vllm/run \
-H 'Content-Type: application/json' \
-H 'Authorization: <JWT TOKEN>' \
--data '{
"prompts": [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is"
]
}'
You should get a response like:
.. code-block:: python
{
"run_id": "52911756-3066-9ae8-bcc9-d9129d1bd262",
"result": {
"result": [
{
"prompt": "Hello, my name is",
"generated_text": " Sarah, and I'm a teacher. I teach elementary school students. One of"
},
{
"prompt": "The president of the United States is",
"generated_text": " elected every four years. This is a democratic system.\n\n5. What"
},
{
"prompt": "The capital of France is",
"generated_text": " Paris.\n"
},
{
"prompt": "The future of AI is",
"generated_text": " bright, but it's important to approach it with a balanced and nuanced perspective."
}
]
},
"run_time_ms": 152.53663063049316
}
You now have an autoscaling endpoint where you only pay for the compute you use!

81
docs/source/serving/deploying_with_docker.md Normal file
View File

@ -0,0 +1,81 @@
(deploying-with-docker)=
# Deploying with Docker
## Use vLLM's Official Docker Image
vLLM offers an official Docker image for deployment.
The image can be used to run OpenAI compatible server and is available on Docker Hub as [vllm/vllm-openai](https://hub.docker.com/r/vllm/vllm-openai/tags).
```console
$ docker run --runtime nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
-p 8000:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model mistralai/Mistral-7B-v0.1
```
```{note}
You can either use the `ipc=host` flag or `--shm-size` flag to allow the
container to access the host's shared memory. vLLM uses PyTorch, which uses shared
memory to share data between processes under the hood, particularly for tensor parallel inference.
```
## Building vLLM's Docker Image from Source
You can build and run vLLM from source via the provided [Dockerfile](https://github.com/vllm-project/vllm/blob/main/Dockerfile). To build vLLM:
```console
$ # optionally specifies: --build-arg max_jobs=8 --build-arg nvcc_threads=2
$ DOCKER_BUILDKIT=1 docker build . --target vllm-openai --tag vllm/vllm-openai
```
```{note}
By default vLLM will build for all GPU types for widest distribution. If you are just building for the
current GPU type the machine is running on, you can add the argument `--build-arg torch_cuda_arch_list=""`
for vLLM to find the current GPU type and build for that.
```
## Building for Arm64/aarch64
A docker container can be built for aarch64 systems such as the Nvidia Grace-Hopper. At time of this writing, this requires the use
of PyTorch Nightly and should be considered **experimental**. Using the flag `--platform "linux/arm64"` will attempt to build for arm64.
```{note}
Multiple modules must be compiled, so this process can take a while. Recommend using `--build-arg max_jobs=` & `--build-arg nvcc_threads=`
flags to speed up build process. However, ensure your `max_jobs` is substantially larger than `nvcc_threads` to get the most benefits.
Keep an eye on memory usage with parallel jobs as it can be substantial (see example below).
```
```console
# Example of building on Nvidia GH200 server. (Memory usage: ~15GB, Build time: ~1475s / ~25 min, Image size: 6.93GB)
$ python3 use_existing_torch.py
$ DOCKER_BUILDKIT=1 docker build . \
--target vllm-openai \
--platform "linux/arm64" \
-t vllm/vllm-gh200-openai:latest \
--build-arg max_jobs=66 \
--build-arg nvcc_threads=2 \
--build-arg torch_cuda_arch_list="9.0+PTX" \
--build-arg vllm_fa_cmake_gpu_arches="90-real"
```
## Use the custom-built vLLM Docker image
To run vLLM with the custom-built Docker image:
```console
$ docker run --runtime nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
-p 8000:8000 \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
vllm/vllm-openai <args...>
```
The argument `vllm/vllm-openai` specifies the image to run, and should be replaced with the name of the custom-built image (the `-t` tag from the build command).
```{note}
**For version 0.4.1 and 0.4.2 only** - the vLLM docker images under these versions are supposed to be run under the root user since a library under the root user's home directory, i.e. `/root/.config/vllm/nccl/cu12/libnccl.so.2.18.1` is required to be loaded during runtime. If you are running the container under a different user, you may need to first change the permissions of the library (and all the parent directories) to allow the user to access it, then run vLLM with environment variable `VLLM_NCCL_SO_PATH=/root/.config/vllm/nccl/cu12/libnccl.so.2.18.1` .
```

88
docs/source/serving/deploying_with_docker.rst
View File

@ -1,88 +0,0 @@
.. _deploying_with_docker:
Deploying with Docker
============================
Use vLLM's Official Docker Image
--------------------------------
vLLM offers an official Docker image for deployment.
The image can be used to run OpenAI compatible server and is available on Docker Hub as `vllm/vllm-openai <https://hub.docker.com/r/vllm/vllm-openai/tags>`_.
.. code-block:: console
$ docker run --runtime nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
-p 8000:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model mistralai/Mistral-7B-v0.1
.. note::
You can either use the ``ipc=host`` flag or ``--shm-size`` flag to allow the
container to access the host's shared memory. vLLM uses PyTorch, which uses shared
memory to share data between processes under the hood, particularly for tensor parallel inference.
Building vLLM's Docker Image from Source
----------------------------------------
You can build and run vLLM from source via the provided `Dockerfile <https://github.com/vllm-project/vllm/blob/main/Dockerfile>`_. To build vLLM:
.. code-block:: console
$ # optionally specifies: --build-arg max_jobs=8 --build-arg nvcc_threads=2
$ DOCKER_BUILDKIT=1 docker build . --target vllm-openai --tag vllm/vllm-openai
.. note::
By default vLLM will build for all GPU types for widest distribution. If you are just building for the
current GPU type the machine is running on, you can add the argument ``--build-arg torch_cuda_arch_list=""``
for vLLM to find the current GPU type and build for that.
Building for Arm64/aarch64
--------------------------
A docker container can be built for aarch64 systems such as the Nvidia Grace-Hopper. At time of this writing, this requires the use
of PyTorch Nightly and should be considered **experimental**. Using the flag ``--platform "linux/arm64"`` will attempt to build for arm64.
.. note::
Multiple modules must be compiled, so this process can take a while. Recommend using ``--build-arg max_jobs=`` & ``--build-arg nvcc_threads=``
flags to speed up build process. However, ensure your ``max_jobs`` is substantially larger than ``nvcc_threads`` to get the most benefits.
Keep an eye on memory usage with parallel jobs as it can be substantial (see example below).
.. code-block:: console
# Example of building on Nvidia GH200 server. (Memory usage: ~15GB, Build time: ~1475s / ~25 min, Image size: 6.93GB)
$ python3 use_existing_torch.py
$ DOCKER_BUILDKIT=1 docker build . \
--target vllm-openai \
--platform "linux/arm64" \
-t vllm/vllm-gh200-openai:latest \
--build-arg max_jobs=66 \
--build-arg nvcc_threads=2 \
--build-arg torch_cuda_arch_list="9.0+PTX" \
--build-arg vllm_fa_cmake_gpu_arches="90-real"
Use the custom-built vLLM Docker image
--------------------------------------
To run vLLM with the custom-built Docker image:
.. code-block:: console
$ docker run --runtime nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
-p 8000:8000 \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
vllm/vllm-openai <args...>
The argument ``vllm/vllm-openai`` specifies the image to run, and should be replaced with the name of the custom-built image (the ``-t`` tag from the build command).
.. note::
**For version 0.4.1 and 0.4.2 only** - the vLLM docker images under these versions are supposed to be run under the root user since a library under the root user's home directory, i.e. ``/root/.config/vllm/nccl/cu12/libnccl.so.2.18.1`` is required to be loaded during runtime. If you are running the container under a different user, you may need to first change the permissions of the library (and all the parent directories) to allow the user to access it, then run vLLM with environment variable ``VLLM_NCCL_SO_PATH=/root/.config/vllm/nccl/cu12/libnccl.so.2.18.1`` .

102
docs/source/serving/deploying_with_dstack.md Normal file
View File

@ -0,0 +1,102 @@
(deploying-with-dstack)=
# Deploying with dstack
```{raw} html
<p align="center">
<img src="https://i.ibb.co/71kx6hW/vllm-dstack.png" alt="vLLM_plus_dstack"/>
</p>
```
vLLM can be run on a cloud based GPU machine with [dstack](https://dstack.ai/), an open-source framework for running LLMs on any cloud. This tutorial assumes that you have already configured credentials, gateway, and GPU quotas on your cloud environment.
To install dstack client, run:
```console
$ pip install "dstack[all]
$ dstack server
```
Next, to configure your dstack project, run:
```console
$ mkdir -p vllm-dstack
$ cd vllm-dstack
$ dstack init
```
Next, to provision a VM instance with LLM of your choice(`NousResearch/Llama-2-7b-chat-hf` for this example), create the following `serve.dstack.yml` file for the dstack `Service`:
```yaml
type: service
python: "3.11"
env:
- MODEL=NousResearch/Llama-2-7b-chat-hf
port: 8000
resources:
gpu: 24GB
commands:
- pip install vllm
- vllm serve $MODEL --port 8000
model:
format: openai
type: chat
name: NousResearch/Llama-2-7b-chat-hf
```
Then, run the following CLI for provisioning:
```console
$ dstack run . -f serve.dstack.yml
⠸ Getting run plan...
Configuration serve.dstack.yml
Project deep-diver-main
User deep-diver
Min resources 2..xCPU, 8GB.., 1xGPU (24GB)
Max price -
Max duration -
Spot policy auto
Retry policy no
# BACKEND REGION INSTANCE RESOURCES SPOT PRICE
1 gcp us-central1 g2-standard-4 4xCPU, 16GB, 1xL4 (24GB), 100GB (disk) yes $0.223804
2 gcp us-east1 g2-standard-4 4xCPU, 16GB, 1xL4 (24GB), 100GB (disk) yes $0.223804
3 gcp us-west1 g2-standard-4 4xCPU, 16GB, 1xL4 (24GB), 100GB (disk) yes $0.223804
...
Shown 3 of 193 offers, $5.876 max
Continue? [y/n]: y
⠙ Submitting run...
⠏ Launching spicy-treefrog-1 (pulling)
spicy-treefrog-1 provisioning completed (running)
Service is published at ...
```
After the provisioning, you can interact with the model by using the OpenAI SDK:
```python
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.<gateway domain>",
api_key="<YOUR-DSTACK-SERVER-ACCESS-TOKEN>"
)
completion = client.chat.completions.create(
model="NousResearch/Llama-2-7b-chat-hf",
messages=[
{
"role": "user",
"content": "Compose a poem that explains the concept of recursion in programming.",
}
]
)
print(completion.choices[0].message.content)
```
```{note}
dstack automatically handles authentication on the gateway using dstack's tokens. Meanwhile, if you don't want to configure a gateway, you can provision dstack `Task` instead of `Service`. The `Task` is for development purpose only. If you want to know more about hands-on materials how to serve vLLM using dstack, check out [this repository](https://github.com/dstackai/dstack-examples/tree/main/deployment/vllm)
```

103
docs/source/serving/deploying_with_dstack.rst
View File

@ -1,103 +0,0 @@
.. _deploying_with_dstack:
Deploying with dstack
============================
.. raw:: html
<p align="center">
<img src="https://i.ibb.co/71kx6hW/vllm-dstack.png" alt="vLLM_plus_dstack"/>
</p>
vLLM can be run on a cloud based GPU machine with `dstack <https://dstack.ai/>`__, an open-source framework for running LLMs on any cloud. This tutorial assumes that you have already configured credentials, gateway, and GPU quotas on your cloud environment.
To install dstack client, run:
.. code-block:: console
$ pip install "dstack[all]
$ dstack server
Next, to configure your dstack project, run:
.. code-block:: console
$ mkdir -p vllm-dstack
$ cd vllm-dstack
$ dstack init
Next, to provision a VM instance with LLM of your choice(`NousResearch/Llama-2-7b-chat-hf` for this example), create the following `serve.dstack.yml` file for the dstack `Service`:
.. code-block:: yaml
type: service
python: "3.11"
env:
- MODEL=NousResearch/Llama-2-7b-chat-hf
port: 8000
resources:
gpu: 24GB
commands:
- pip install vllm
- vllm serve $MODEL --port 8000
model:
format: openai
type: chat
name: NousResearch/Llama-2-7b-chat-hf
Then, run the following CLI for provisioning:
.. code-block:: console
$ dstack run . -f serve.dstack.yml
⠸ Getting run plan...
Configuration serve.dstack.yml
Project deep-diver-main
User deep-diver
Min resources 2..xCPU, 8GB.., 1xGPU (24GB)
Max price -
Max duration -
Spot policy auto
Retry policy no
# BACKEND REGION INSTANCE RESOURCES SPOT PRICE
1 gcp us-central1 g2-standard-4 4xCPU, 16GB, 1xL4 (24GB), 100GB (disk) yes $0.223804
2 gcp us-east1 g2-standard-4 4xCPU, 16GB, 1xL4 (24GB), 100GB (disk) yes $0.223804
3 gcp us-west1 g2-standard-4 4xCPU, 16GB, 1xL4 (24GB), 100GB (disk) yes $0.223804
...
Shown 3 of 193 offers, $5.876 max
Continue? [y/n]: y
⠙ Submitting run...
⠏ Launching spicy-treefrog-1 (pulling)
spicy-treefrog-1 provisioning completed (running)
Service is published at ...
After the provisioning, you can interact with the model by using the OpenAI SDK:
.. code-block:: python
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.<gateway domain>",
api_key="<YOUR-DSTACK-SERVER-ACCESS-TOKEN>"
)
completion = client.chat.completions.create(
model="NousResearch/Llama-2-7b-chat-hf",
messages=[
{
"role": "user",
"content": "Compose a poem that explains the concept of recursion in programming.",
}
]
)
print(completion.choices[0].message.content)
.. note::
dstack automatically handles authentication on the gateway using dstack's tokens. Meanwhile, if you don't want to configure a gateway, you can provision dstack `Task` instead of `Service`. The `Task` is for development purpose only. If you want to know more about hands-on materials how to serve vLLM using dstack, check out `this repository <https://github.com/dstackai/dstack-examples/tree/main/deployment/vllm>`__

44
docs/source/serving/deploying_with_helm.rst → docs/source/serving/deploying_with_helm.md
View File

@ -1,7 +1,6 @@
.. _deploying_with_helm:
(deploying-with-helm)=
Deploying with Helm
===================
# Deploying with Helm
A Helm chart to deploy vLLM for Kubernetes
@ -9,44 +8,42 @@ Helm is a package manager for Kubernetes. It will help you to deploy vLLM on k8s
This guide will walk you through the process of deploying vLLM with Helm, including the necessary prerequisites, steps for helm install and documentation on architecture and values file.
Prerequisites
-------------
## Prerequisites
Before you begin, ensure that you have the following:
- A running Kubernetes cluster
- NVIDIA Kubernetes Device Plugin (``k8s-device-plugin``): This can be found at `https://github.com/NVIDIA/k8s-device-plugin <https://github.com/NVIDIA/k8s-device-plugin>`__
- NVIDIA Kubernetes Device Plugin (`k8s-device-plugin`): This can be found at [https://github.com/NVIDIA/k8s-device-plugin](https://github.com/NVIDIA/k8s-device-plugin)
- Available GPU resources in your cluster
- S3 with the model which will be deployed
Installing the chart
--------------------
## Installing the chart
To install the chart with the release name ``test-vllm``:
To install the chart with the release name `test-vllm`:
.. code-block:: console
```console
helm upgrade --install --create-namespace --namespace=ns-vllm test-vllm . -f values.yaml --set secrets.s3endpoint=$ACCESS_POINT --set secrets.s3bucketname=$BUCKET --set secrets.s3accesskeyid=$ACCESS_KEY --set secrets.s3accesskey=$SECRET_KEY
```
helm upgrade --install --create-namespace --namespace=ns-vllm test-vllm . -f values.yaml --set secrets.s3endpoint=$ACCESS_POINT --set secrets.s3bucketname=$BUCKET --set secrets.s3accesskeyid=$ACCESS_KEY --set secrets.s3accesskey=$SECRET_KEY
## Uninstalling the Chart
Uninstalling the Chart
----------------------
To uninstall the `test-vllm` deployment:
To uninstall the ``test-vllm`` deployment:
.. code-block:: console
helm uninstall test-vllm --namespace=ns-vllm
```console
helm uninstall test-vllm --namespace=ns-vllm
```
The command removes all the Kubernetes components associated with the
chart **including persistent volumes** and deletes the release.
Architecture
------------
## Architecture
.. image:: architecture_helm_deployment.png
```{image} architecture_helm_deployment.png
```
Values
------
## Values
```{eval-rst}
.. list-table:: Values
:widths: 25 25 25 25
:header-rows: 1
@ -251,3 +248,4 @@ Values
- string
- test
- Release name
```

171
docs/source/serving/deploying_with_k8s.md Normal file
View File

@ -0,0 +1,171 @@
(deploying-with-k8s)=
# Deploying with Kubernetes
Using Kubernetes to deploy vLLM is a scalable and efficient way to serve machine learning models. This guide will walk you through the process of deploying vLLM with Kubernetes, including the necessary prerequisites, steps for deployment, and testing.
## Prerequisites
Before you begin, ensure that you have the following:
- A running Kubernetes cluster
- NVIDIA Kubernetes Device Plugin (`k8s-device-plugin`): This can be found at `https://github.com/NVIDIA/k8s-device-plugin/`
- Available GPU resources in your cluster
## Deployment Steps
1. **Create a PVC , Secret and Deployment for vLLM**
PVC is used to store the model cache and it is optional, you can use hostPath or other storage options
```yaml
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: mistral-7b
namespace: default
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 50Gi
storageClassName: default
volumeMode: Filesystem
```
Secret is optional and only required for accessing gated models, you can skip this step if you are not using gated models
```yaml
apiVersion: v1
kind: Secret
metadata:
name: hf-token-secret
namespace: default
type: Opaque
data:
token: "REPLACE_WITH_TOKEN"
```
Create a deployment file for vLLM to run the model server. The following example deploys the `Mistral-7B-Instruct-v0.3` model:
```yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: mistral-7b
namespace: default
labels:
app: mistral-7b
spec:
replicas: 1
selector:
matchLabels:
app: mistral-7b
template:
metadata:
labels:
app: mistral-7b
spec:
volumes:
- name: cache-volume
persistentVolumeClaim:
claimName: mistral-7b
# vLLM needs to access the host's shared memory for tensor parallel inference.
- name: shm
emptyDir:
medium: Memory
sizeLimit: "2Gi"
containers:
- name: mistral-7b
image: vllm/vllm-openai:latest
command: ["/bin/sh", "-c"]
args: [
"vllm serve mistralai/Mistral-7B-Instruct-v0.3 --trust-remote-code --enable-chunked-prefill --max_num_batched_tokens 1024"
]
env:
- name: HUGGING_FACE_HUB_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
ports:
- containerPort: 8000
resources:
limits:
cpu: "10"
memory: 20G
nvidia.com/gpu: "1"
requests:
cpu: "2"
memory: 6G
nvidia.com/gpu: "1"
volumeMounts:
- mountPath: /root/.cache/huggingface
name: cache-volume
- name: shm
mountPath: /dev/shm
livenessProbe:
httpGet:
path: /health
port: 8000
initialDelaySeconds: 60
periodSeconds: 10
readinessProbe:
httpGet:
path: /health
port: 8000
initialDelaySeconds: 60
periodSeconds: 5
```
2. **Create a Kubernetes Service for vLLM**
Next, create a Kubernetes Service file to expose the `mistral-7b` deployment:
```yaml
apiVersion: v1
kind: Service
metadata:
name: mistral-7b
namespace: default
spec:
ports:
- name: http-mistral-7b
port: 80
protocol: TCP
targetPort: 8000
# The label selector should match the deployment labels & it is useful for prefix caching feature
selector:
app: mistral-7b
sessionAffinity: None
type: ClusterIP
```
3. **Deploy and Test**
Apply the deployment and service configurations using `kubectl apply -f <filename>`:
```console
kubectl apply -f deployment.yaml
kubectl apply -f service.yaml
```
To test the deployment, run the following `curl` command:
```console
curl http://mistral-7b.default.svc.cluster.local/v1/completions \
-H "Content-Type: application/json" \
-d '{
"model": "mistralai/Mistral-7B-Instruct-v0.3",
"prompt": "San Francisco is a",
"max_tokens": 7,
"temperature": 0
}'
```
If the service is correctly deployed, you should receive a response from the vLLM model.
## Conclusion
Deploying vLLM with Kubernetes allows for efficient scaling and management of ML models leveraging GPU resources. By following the steps outlined above, you should be able to set up and test a vLLM deployment within your Kubernetes cluster. If you encounter any issues or have suggestions, please feel free to contribute to the documentation.

Some files were not shown because too many files have changed in this diff Show More