xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-20 06:07:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Logging] Improve log for when DeepEP HT disables CUDA Graphs (#25531)

Browse Source 
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
This commit is contained in:
Tyler Michael Smith 2025-09-24 18:43:06 -04:00 committed by GitHub
parent e6750d0b18
commit fea8006062
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/platforms/cuda.py
View File

@ -186,11 +186,12 @@ class CudaPlatformBase(Platform):
# if torch compile cache key issue fixed # if torch compile cache key issue fixed
# See https://github.com/vllm-project/vllm/pull/25093 # See https://github.com/vllm-project/vllm/pull/25093
logger.info( logger.info(
"Data Parallel: disabling cudagraphs since DP " "WideEP: Disabling CUDA Graphs since DeepEP high-throughput "
"with DeepEP high-throughput kernels are not CUDA Graph " "kernels are optimized for prefill and are incompatible with "
"compatible. The DeepEP low-latency kernels are CUDA Graph " "CUDA Graphs. "
"compatible. Set the all_to_all backend to deepep_low_latency " "In order to use CUDA Graphs for decode-optimized workloads, "
"to use those kernels instead.") "set VLLM_ALL2ALL_BACKEND to another option, such as "
"deepep_low_latency, pplx, or allgather_reducescatter.")
compilation_config.cudagraph_mode = CUDAGraphMode.NONE compilation_config.cudagraph_mode = CUDAGraphMode.NONE
@classmethod @classmethod