xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 10:18:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Doc] Update help text for --distributed-executor-backend (#10231)

Browse Source 
Signed-off-by: Russell Bryant <rbryant@redhat.com>
This commit is contained in:
Russell Bryant 2024-11-11 20:55:09 -05:00 committed by GitHub
parent d1c6799b88
commit 9cdba9669c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 14 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/config.py
View File

@ -951,9 +951,12 @@ class ParallelConfig:
https://docs.ray.io/en/latest/ray-observability/user-guides/profiling.html#profiling-nsight-profiler.
placement_group: ray distributed model workers placement group.
distributed_executor_backend: Backend to use for distributed model
workers, either "ray" or "mp" (multiprocessing). If either
pipeline_parallel_size or tensor_parallel_size is greater than 1,
will default to "ray" if Ray is installed or "mp" otherwise.
workers, either "ray" or "mp" (multiprocessing). If the product
of pipeline_parallel_size and tensor_parallel_size is less than
or equal to the number of GPUs available, "mp" will be used to
keep processing on a single host. Otherwise, this will default
to "ray" if Ray is installed and fail otherwise. Note that tpu
and hpu only support Ray for distributed inference.
"""
def __init__(

11
vllm/engine/arg_utils.py
View File

@ -369,9 +369,14 @@ class EngineArgs:
'--distributed-executor-backend',
choices=['ray', 'mp'],
default=EngineArgs.distributed_executor_backend,
help='Backend to use for distributed serving. When more than 1 GPU '
'is used, will be automatically set to "ray" if installed '
'or "mp" (multiprocessing) otherwise.')
help='Backend to use for distributed model '
'workers, either "ray" or "mp" (multiprocessing). If the product '
'of pipeline_parallel_size and tensor_parallel_size is less than '
'or equal to the number of GPUs available, "mp" will be used to '
'keep processing on a single host. Otherwise, this will default '
'to "ray" if Ray is installed and fail otherwise. Note that tpu '
'and hpu only support Ray for distributed inference.')
parser.add_argument(
'--worker-use-ray',
action='store_true',