Skip to content

Commit

Permalink
[Doc] Update help text for --distributed-executor-backend (#10231)
Browse files Browse the repository at this point in the history
Signed-off-by: Russell Bryant <[email protected]>
  • Loading branch information
russellb authored Nov 12, 2024
1 parent d1c6799 commit 9cdba96
Show file tree
Hide file tree
Showing 2 changed files with 14 additions and 6 deletions.
9 changes: 6 additions & 3 deletions vllm/config.py
Original file line number Diff line number Diff line change
Expand Up @@ -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__(
Expand Down
11 changes: 8 additions & 3 deletions vllm/engine/arg_utils.py
Original file line number Diff line number Diff line change
Expand Up @@ -369,9 +369,14 @@ def add_cli_args(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
'--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',
Expand Down

0 comments on commit 9cdba96

Please sign in to comment.