frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

disable flashinfer warmup

Browse Source
This commit is contained in:
Woosuk Kwon
2025-10-16 16:49:29 +00:00
parent 8935ca208d
commit 69c9a01538
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/warmup/kernel_warmup.py
View File

@ -36,9 +36,9 @@ def kernel_warmup(worker: "Worker"):
max_tokens = worker.scheduler_config.max_num_batched_tokens max_tokens = worker.scheduler_config.max_num_batched_tokens
deep_gemm_warmup(model, max_tokens) deep_gemm_warmup(model, max_tokens)
# FlashInfer autotune for Hopper (SM 9.0) and Blackwell (SM 10.0) GPUs # # FlashInfer autotune for Hopper (SM 9.0) and Blackwell (SM 10.0) GPUs
if has_flashinfer() and current_platform.has_device_capability(90): # if has_flashinfer() and current_platform.has_device_capability(90):
flashinfer_autotune(worker.model_runner) # flashinfer_autotune(worker.model_runner)
# FlashInfer attention warmup # FlashInfer attention warmup
# Only warmup if the model has FlashInfer attention groups # Only warmup if the model has FlashInfer attention groups