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

[Bugfix] Properly set distributed_executor_backend in ParallelConfig (#4816)

Browse Source
This commit is contained in:
zifeitong 2024-05-15 07:22:09 -07:00 committed by GitHub
parent e9cdd2b1e2
commit a5675d348b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 8 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
vllm/config.py
View File

@ -531,6 +531,7 @@ class ParallelConfig:
If None, will use synchronous tokenization. If None, will use synchronous tokenization.
ray_workers_use_nsight: Whether to profile Ray workers with nsight, see ray_workers_use_nsight: Whether to profile Ray workers with nsight, see
https://docs.ray.io/en/latest/ray-observability/user-guides/profiling.html#profiling-nsight-profiler. 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 distributed_executor_backend: Backend to use for distributed model
workers, either "ray" or "mp" (multiprocessing). If either workers, either "ray" or "mp" (multiprocessing). If either
pipeline_parallel_size or tensor_parallel_size is greater than 1, pipeline_parallel_size or tensor_parallel_size is greater than 1,

10
vllm/engine/arg_utils.py
View File

@ -548,14 +548,18 @@ class EngineArgs:
model_config.get_sliding_window(), model_config.get_sliding_window(),
self.enable_prefix_caching) self.enable_prefix_caching)
parallel_config = ParallelConfig( parallel_config = ParallelConfig(
self.pipeline_parallel_size, self.tensor_parallel_size, self.pipeline_parallel_size,
self.worker_use_ray, self.max_parallel_loading_workers, self.tensor_parallel_size,
self.worker_use_ray,
self.max_parallel_loading_workers,
self.disable_custom_all_reduce, self.disable_custom_all_reduce,
TokenizerPoolConfig.create_config( TokenizerPoolConfig.create_config(
self.tokenizer_pool_size, self.tokenizer_pool_size,
self.tokenizer_pool_type, self.tokenizer_pool_type,
self.tokenizer_pool_extra_config, self.tokenizer_pool_extra_config,
), self.ray_workers_use_nsight) ),
self.ray_workers_use_nsight,
distributed_executor_backend=self.distributed_executor_backend)
speculative_config = SpeculativeConfig.maybe_create_spec_config( speculative_config = SpeculativeConfig.maybe_create_spec_config(
target_model_config=model_config, target_model_config=model_config,