e489ad7a21
- **Add SPDX license headers to python source files**
- **Check for SPDX headers using pre-commit**
commit 9d7ef44c3cfb72ca4c32e1c677d99259d10d4745
Author: Russell Bryant <rbryant@redhat.com>
Date: Fri Jan 31 14:18:24 2025 -0500
Add SPDX license headers to python source files
This commit adds SPDX license headers to python source files as
recommended to
the project by the Linux Foundation. These headers provide a concise way
that is
both human and machine readable for communicating license information
for each
source file. It helps avoid any ambiguity about the license of the code
and can
also be easily used by tools to help manage license compliance.
The Linux Foundation runs license scans against the codebase to help
ensure
we are in compliance with the licenses of the code we use, including
dependencies. Having these headers in place helps that tool do its job.
More information can be found on the SPDX site:
- https://spdx.dev/learn/handling-license-info/
Signed-off-by: Russell Bryant <rbryant@redhat.com>
commit 5a1cf1cb3b80759131c73f6a9dddebccac039dea
Author: Russell Bryant <rbryant@redhat.com>
Date: Fri Jan 31 14:36:32 2025 -0500
Check for SPDX headers using pre-commit
Signed-off-by: Russell Bryant <rbryant@redhat.com>
---------
Signed-off-by: Russell Bryant <rbryant@redhat.com>
68 lines
2.4 KiB
Python
68 lines
2.4 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
|
|
import torch
|
|
|
|
from vllm.engine.arg_utils import EngineArgs
|
|
from vllm.utils import get_distributed_init_method, get_ip, get_open_port
|
|
from vllm.worker.cache_engine import CacheEngine
|
|
from vllm.worker.worker import Worker
|
|
|
|
|
|
def test_gpu_memory_profiling():
|
|
# Tests the gpu profiling that happens in order to determine the number of
|
|
# KV cache blocks that we can allocate on the GPU.
|
|
# This test mocks the maximum available gpu memory so that it can run on
|
|
# any gpu setup.
|
|
|
|
# Set up engine args to build a worker.
|
|
engine_args = EngineArgs(model="facebook/opt-125m",
|
|
dtype="half",
|
|
load_format="dummy")
|
|
engine_config = engine_args.create_engine_config()
|
|
engine_config.cache_config.num_gpu_blocks = 1000
|
|
engine_config.cache_config.num_cpu_blocks = 1000
|
|
|
|
# Create the worker.
|
|
distributed_init_method = get_distributed_init_method(
|
|
get_ip(), get_open_port())
|
|
worker = Worker(
|
|
vllm_config=engine_config,
|
|
local_rank=0,
|
|
rank=0,
|
|
distributed_init_method=distributed_init_method,
|
|
is_driver_worker=True,
|
|
)
|
|
|
|
# Set 10GiB as the total gpu ram to be device-agnostic
|
|
def mock_mem_info():
|
|
current_usage = torch.cuda.memory_stats(
|
|
)["allocated_bytes.all.current"]
|
|
mock_total_bytes = 10 * 1024**3
|
|
free = mock_total_bytes - current_usage
|
|
|
|
return (free, mock_total_bytes)
|
|
|
|
from unittest.mock import patch
|
|
with patch("torch.cuda.mem_get_info", side_effect=mock_mem_info):
|
|
# Load the model so we can profile it
|
|
worker.init_device()
|
|
worker.load_model()
|
|
gpu_blocks, _ = worker.determine_num_available_blocks()
|
|
|
|
# Peak vram usage by torch should be 0.47 GiB
|
|
# Model weights take 0.25 GiB
|
|
# No memory should be allocated outside of torch
|
|
# 9.0 GiB should be the utilization target
|
|
# 8.28 GiB should be available for the KV cache
|
|
block_size = CacheEngine.get_cache_block_size(
|
|
engine_config.cache_config, engine_config.model_config,
|
|
engine_config.parallel_config)
|
|
|
|
expected_blocks = (8.28 * 1024**3) // block_size
|
|
|
|
# Check within a small tolerance for portability
|
|
# Hardware, kernel, or dependency changes could all affect memory
|
|
# utilization.
|
|
# A 100 block tolerance here should be about 60MB of wiggle room.
|
|
assert abs(gpu_blocks - expected_blocks) < 100
|