2e9a2227ec
Currently we need to call rotary embedding kernel for each LoRA, which makes it hard to serve multiple long context length LoRA. Add batched rotary embedding kernel and pipe it through. It replaces the rotary embedding layer to the one that is aware of multiple cos-sin-cache per scaling factors. Follow up of https://github.com/vllm-project/vllm/pull/3095/files
68 lines
1.3 KiB
TOML
68 lines
1.3 KiB
TOML
[build-system]
|
|
# Should be mirrored in requirements-build.txt
|
|
requires = [
|
|
"cmake>=3.21",
|
|
"ninja",
|
|
"packaging",
|
|
"setuptools >= 49.4.0",
|
|
"torch == 2.3.0",
|
|
"wheel",
|
|
]
|
|
build-backend = "setuptools.build_meta"
|
|
|
|
[tool.ruff]
|
|
# Allow lines to be as long as 80.
|
|
line-length = 80
|
|
exclude = [
|
|
# External file, leaving license intact
|
|
"examples/fp8/quantizer/quantize.py"
|
|
]
|
|
|
|
[tool.ruff.lint]
|
|
select = [
|
|
# pycodestyle
|
|
"E",
|
|
# Pyflakes
|
|
"F",
|
|
# pyupgrade
|
|
# "UP",
|
|
# flake8-bugbear
|
|
"B",
|
|
# flake8-simplify
|
|
"SIM",
|
|
# isort
|
|
# "I",
|
|
"G",
|
|
]
|
|
ignore = [
|
|
# star imports
|
|
"F405", "F403",
|
|
# lambda expression assignment
|
|
"E731",
|
|
# Loop control variable not used within loop body
|
|
"B007",
|
|
]
|
|
|
|
[tool.mypy]
|
|
python_version = "3.8"
|
|
|
|
ignore_missing_imports = true
|
|
check_untyped_defs = true
|
|
follow_imports = "skip"
|
|
|
|
files = "vllm"
|
|
# TODO(woosuk): Include the code from Megatron and HuggingFace.
|
|
exclude = [
|
|
"vllm/model_executor/parallel_utils/|vllm/model_executor/models/",
|
|
# Ignore triton kernels in ops.
|
|
'vllm/attention/ops/.*\.py$'
|
|
]
|
|
|
|
[tool.codespell]
|
|
ignore-words-list = "dout, te, indicies"
|
|
skip = "./tests/prompts,./benchmarks/sonnet.txt,./tests/lora/data"
|
|
|
|
[tool.isort]
|
|
use_parentheses = true
|
|
skip_gitignore = true
|