mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2025-12-15 13:55:38 +08:00
[Log] Warning Once for Cutlass MLA (#23137)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
parent
e75f342261
commit
90bbe0a5ad
@ -115,7 +115,7 @@ class CutlassMLAImpl(MLACommonImpl[MLACommonMetadata]):
|
|||||||
self._use_old_cutlass_mla = False
|
self._use_old_cutlass_mla = False
|
||||||
force_old_cutlass = os.environ.get("FORCE_OLD_CUTLASS_MLA", None)
|
force_old_cutlass = os.environ.get("FORCE_OLD_CUTLASS_MLA", None)
|
||||||
if force_old_cutlass:
|
if force_old_cutlass:
|
||||||
logger.warning("Forcing old cutlass mla kernel")
|
logger.warning_once("Forcing old cutlass mla kernel")
|
||||||
self._use_old_cutlass_mla = True
|
self._use_old_cutlass_mla = True
|
||||||
|
|
||||||
# TODO: Currently, num_kv_splits is limited to 16 to avoid hanging
|
# TODO: Currently, num_kv_splits is limited to 16 to avoid hanging
|
||||||
@ -123,8 +123,8 @@ class CutlassMLAImpl(MLACommonImpl[MLACommonMetadata]):
|
|||||||
# FORCE_NUM_KV_SPLITS=1
|
# FORCE_NUM_KV_SPLITS=1
|
||||||
force_num_kv_splits = os.environ.get("FORCE_NUM_KV_SPLITS", None)
|
force_num_kv_splits = os.environ.get("FORCE_NUM_KV_SPLITS", None)
|
||||||
if force_num_kv_splits:
|
if force_num_kv_splits:
|
||||||
logger.warning("Forcing num_kv_splits to %d",
|
logger.warning_once("Forcing num_kv_splits to %d",
|
||||||
int(force_num_kv_splits))
|
int(force_num_kv_splits))
|
||||||
self._num_kv_splits = int(force_num_kv_splits)
|
self._num_kv_splits = int(force_num_kv_splits)
|
||||||
else:
|
else:
|
||||||
self._num_kv_splits = -1 # => Auto-detect
|
self._num_kv_splits = -1 # => Auto-detect
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user