[CPU] Skip Triton kernel monkey-patches when Triton-CPU is available (#44991)
Signed-off-by: jmamou <jonathan.mamou@intel.com> Co-authored-by: Li, Jiang <jiang1.li@intel.com>
This commit is contained in:
co-authored by
Li, Jiang <jiang1.li@intel.com>
parent
7299e6509e
commit
351c72d6e5
@@ -60,6 +60,15 @@ class CPUModelRunner(GPUModelRunner):
|
||||
v.gpu = v.cpu
|
||||
|
||||
def _postprocess_triton(self) -> None:
|
||||
from vllm.triton_utils import HAS_TRITON
|
||||
|
||||
if HAS_TRITON:
|
||||
logger.info(
|
||||
"Triton-CPU backend is available; skipping C++ monkey-patches "
|
||||
"for Triton kernels."
|
||||
)
|
||||
return
|
||||
|
||||
import vllm.v1.worker.block_table
|
||||
|
||||
vllm.v1.worker.block_table._compute_slot_mapping_kernel = (
|
||||
|
||||
Reference in New Issue
Block a user