xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-20 06:37:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Support MLA for CompressedTensorsWNA16 (#13725)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin 2025-02-25 01:10:31 -05:00 committed by GitHub
parent 4a8cfc7551
commit 18e505930d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/attention/backends/mla/common.py
View File

@ -1130,13 +1130,13 @@ class MLACommonImpl(MLAAttentionImpl[T], Generic[T]):
) )
def get_layer_weight(layer): def get_layer_weight(layer):
if hasattr(layer, "weight"): WEIGHT_NAMES = ("weight", "qweight", "weight_packed")
return layer.weight for attr in WEIGHT_NAMES:
elif hasattr(layer, "qweight"): if hasattr(layer, attr):
return layer.qweight return getattr(layer, attr)
else: raise AttributeError(
raise AttributeError( f"Layer '{layer}' has no recognized weight attribute:"
f"Layer '{layer}' has neither weight nor qweight") f" {WEIGHT_NAMES}.")
def get_and_maybe_dequant_weights(layer: LinearBase): def get_and_maybe_dequant_weights(layer: LinearBase):
if not isinstance(layer.quant_method, UnquantizedLinearMethod): if not isinstance(layer.quant_method, UnquantizedLinearMethod):