xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-13 20:44:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Significant performance drop on CPUs with --num-scheduler-steps > 1 (#11794)

Browse Source
This commit is contained in:
Jie Fu (傅杰) 2025-01-08 08:15:50 +08:00 committed by GitHub
parent 973f5dc581
commit a4e2b26856
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/engine/arg_utils.py
View File

@ -1157,6 +1157,12 @@ class EngineArgs:
if self.enable_chunked_prefill and self.pipeline_parallel_size > 1: if self.enable_chunked_prefill and self.pipeline_parallel_size > 1:
raise ValueError("Multi-Step Chunked-Prefill is not supported " raise ValueError("Multi-Step Chunked-Prefill is not supported "
"for pipeline-parallel-size > 1") "for pipeline-parallel-size > 1")
from vllm.platforms import current_platform
if current_platform.is_cpu():
logger.warning("Multi-Step (--num-scheduler-steps > 1) is "
"currently not supported for CPUs and has been "
"disabled.")
self.num_scheduler_steps = 1
# make sure num_lookahead_slots is set the higher value depending on # make sure num_lookahead_slots is set the higher value depending on
# if we are using speculative decoding or multi-step # if we are using speculative decoding or multi-step