Fix FBGEMM integration (#18002)
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -63,7 +63,9 @@ class FBGEMMFp8Config(QuantizationConfig):
|
||||
def get_quant_method(self, layer: torch.nn.Module,
|
||||
prefix: str) -> Optional["QuantizeMethodBase"]:
|
||||
if isinstance(layer, LinearBase):
|
||||
if is_layer_skipped(prefix, self.ignore_list):
|
||||
if is_layer_skipped(prefix=prefix,
|
||||
ignored_layers=self.ignore_list,
|
||||
fused_mapping=self.packed_modules_mapping):
|
||||
return UnquantizedLinearMethod()
|
||||
return FBGEMMFp8LinearMethod(self)
|
||||
return None
|
||||
|
||||
Reference in New Issue
Block a user