biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] DeepEPHighThroughtput - Enable Inductor pass (#21311)

Browse Source 
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
This commit is contained in:
Varun Sundar Rabindranath
2025-07-22 12:05:45 +05:30
committed by GitHub
parent c17231e827
commit 8425f785ad
1 changed files with 0 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/platforms/cuda.py
View File

@@ -182,9 +182,6 @@ class CudaPlatformBase(Platform):
compilation_config.use_cudagraph = False compilation_config.use_cudagraph = False
if model_config is not None: if model_config is not None:
model_config.enforce_eager = True model_config.enforce_eager = True
# TODO (varun): Turning this ON gives incorrect results for the
# Deepseek-V2-lite model.
vllm_config.compilation_config.use_inductor = False
@classmethod @classmethod
def get_current_memory_usage(cls, def get_current_memory_usage(cls,