[Model] add GLM-5.3-Flash support (#53906)

This commit is contained in:
Jiangyun Zhu
2026-09-03 09:40:35 -07:00
committed by GitHub
parent da8ec28268
commit 98ed0856f3
94 changed files with 17927 additions and 266 deletions
+26
View File
@@ -292,6 +292,32 @@ def has_flashinfer_moe() -> bool:
)
@functools.cache
def has_flashinfer_sm90_nope_mla() -> bool:
"""FlashInfer SM90 NoPE MLA (FP8 KV with in-kernel dequant, kpe=0).
Feature-detected via the ``ckv_scale_arr`` run() kwarg introduced with
the SM90 NoPE support (FlashInfer >= 0.6.18), so dev builds carry the
gate without a version parse.
"""
if not has_flashinfer():
return False
try:
import inspect
from flashinfer.mla import BatchMLAPagedAttentionWrapper
except ImportError:
return False
try:
params = inspect.signature(BatchMLAPagedAttentionWrapper.run).parameters
except (TypeError, ValueError):
return False
return (
"ckv_scale_arr" in params
and params["ckv_scale_arr"].kind is inspect.Parameter.KEYWORD_ONLY
)
@functools.cache
def has_flashinfer_sparse_mla_sm120() -> bool:
"""Return ``True`` if FlashInfer sparse MLA decode support is available."""