xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 05:34:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

[misc] Clean up ray compiled graph type hints (#13731)

Browse Source
This commit is contained in:
Rui Qiao 2025-02-25 00:37:08 -08:00 committed by GitHub
parent 2d87d7d1ac
commit 3173c3b34e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 17 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
vllm/executor/ray_distributed_executor.py
View File

@ -528,10 +528,18 @@ class RayDistributedExecutor(DistributedExecutorBase):
envs.VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM) envs.VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM)
with InputNode() as input_data: with InputNode() as input_data:
# Example DAG: PP=2, TP=4 # Example DAG: PP=2, TP=4
# (ExecuteModelReq, None) -> 0 -> (ExecuteModelReq, IntermediateOutput) -> 4 -> SamplerOutput # noqa: E501 #
# -> 1 -> (ExecuteModelReq, IntermediateOutput) -> 5 -> SamplerOutput # noqa: E501 # For V0:
# -> 2 -> (ExecuteModelReq, IntermediateOutput) -> 6 -> SamplerOutput # noqa: E501 # ExecuteModelRequest -> 0 -> (ExecuteModelReq, IntermediateTensors) -> 4 -> SamplerOutput # noqa: E501
# -> 3 -> (ExecuteModelReq, IntermediateOutput) -> 7 -> SamplerOutput # noqa: E501 # ExecuteModelRequest -> 1 -> (ExecuteModelReq, IntermediateTensors) -> 5 -> SamplerOutput # noqa: E501
# ExecuteModelRequest -> 2 -> (ExecuteModelReq, IntermediateTensors) -> 6 -> SamplerOutput # noqa: E501
# ExecuteModelRequest -> 3 -> (ExecuteModelReq, IntermediateTensors) -> 7 -> SamplerOutput # noqa: E501
#
# For V1:
# SchedulerOutput -> 0 -> (SchedulerOutput, IntermediateTensors) -> 4 -> ModelRunnerOutput # noqa: E501
# SchedulerOutput -> 1 -> (SchedulerOutput, IntermediateTensors) -> 5 -> ModelRunnerOutput # noqa: E501
# SchedulerOutput -> 2 -> (SchedulerOutput, IntermediateTensors) -> 6 -> ModelRunnerOutput # noqa: E501
# SchedulerOutput -> 3 -> (SchedulerOutput, IntermediateTensors) -> 7 -> ModelRunnerOutput # noqa: E501
# All workers in the first TP group will take in the # All workers in the first TP group will take in the
# ExecuteModelRequest as input. # ExecuteModelRequest as input.

7
vllm/executor/ray_utils.py
View File

@ -114,8 +114,11 @@ try:
def execute_model_ray( def execute_model_ray(
self, self,
scheduler_output: "SchedulerOutput", scheduler_output: Union["SchedulerOutput",
) -> "ModelRunnerOutput": Tuple["SchedulerOutput",
"IntermediateTensors"]],
) -> Union["ModelRunnerOutput", Tuple["SchedulerOutput",
"IntermediateTensors"]]:
# this method is used to compile ray CG, # this method is used to compile ray CG,
# and it needs a special logic of self.setup_device_if_necessary() # and it needs a special logic of self.setup_device_if_necessary()
self.setup_device_if_necessary() self.setup_device_if_necessary()