xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 09:55:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

align llm_engine and async_engine. (#1081)

Browse Source
This commit is contained in:
Roy 2023-09-19 02:49:10 +08:00 committed by GitHub
parent fbe66e1d0b
commit 95592fa00a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/engine/async_llm_engine.py
View File

@ -183,10 +183,9 @@ class _AsyncLLMEngine(LLMEngine):
and updates the scheduler with the model outputs. Finally, it decodes
the sequences and returns the newly generated results.
"""
(seq_group_metadata_list, scheduler_outputs,
early_return) = self._schedule()
if early_return is not None:
return early_return
seq_group_metadata_list, scheduler_outputs, ignored = self._schedule()
if scheduler_outputs.is_empty():
return ignored
# Execute the model.
output = await self._run_workers_async(
@ -197,7 +196,7 @@ class _AsyncLLMEngine(LLMEngine):
blocks_to_copy=scheduler_outputs.blocks_to_copy,
)
return self._process_model_outputs(output, scheduler_outputs)
return self._process_model_outputs(output, scheduler_outputs) + ignored
async def _run_workers_async(
self,