[Kernel] Add KernelConfig flag to enable/disable FlashInfer autotune (#34006)
Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com> Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
co-authored by
Luka Govedič
parent
edb359cce4
commit
dd6a6e1190
@@ -36,8 +36,13 @@ def kernel_warmup(worker: "Worker"):
|
||||
max_tokens = worker.scheduler_config.max_num_batched_tokens
|
||||
deep_gemm_warmup(model, max_tokens)
|
||||
|
||||
enable_flashinfer_autotune = (
|
||||
worker.vllm_config.kernel_config.enable_flashinfer_autotune
|
||||
)
|
||||
# FlashInfer autotune for Hopper (SM 9.0) and Blackwell (SM 10.0) GPUs
|
||||
if has_flashinfer() and current_platform.has_device_capability(90):
|
||||
if enable_flashinfer_autotune is False:
|
||||
logger.info("Skipping FlashInfer autotune because it is disabled.")
|
||||
elif has_flashinfer() and current_platform.has_device_capability(90):
|
||||
flashinfer_autotune(worker.model_runner)
|
||||
|
||||
# FlashInfer attention warmup
|
||||
|
||||
Reference in New Issue
Block a user