biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Fix missing is_layer_skipped check for FusedMoE in AWQConfig (#32935)

Browse Source 
Signed-off-by: jon <joninco@bullpoint.org>
This commit is contained in:
joninco
2026-01-23 17:19:56 -05:00
committed by GitHub
parent 37c9859fab
commit ebd0a17e0e
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/layers/quantization/awq.py
View File

@@ -106,7 +106,7 @@ class AWQConfig(QuantizationConfig):
return AWQLinearMethod(self)
elif isinstance(layer, FusedMoE):
# Lazy import to avoid circular import.
from .awq_marlin import AWQMarlinConfig, AWQMarlinMoEMethod
from .awq_marlin import AWQMarlinConfig
from .moe_wna16 import MoeWNA16Config
from .utils.marlin_utils import check_moe_marlin_supports_layer
@@ -121,6 +121,7 @@ class AWQConfig(QuantizationConfig):
"group_size": self.group_size,
"zero_point": self.zero_point,
"lm_head": False,
"modules_to_not_convert": self.modules_to_not_convert,
}
return MoeWNA16Config.from_config(config).get_quant_method(
layer, prefix
@@ -136,7 +137,7 @@ class AWQConfig(QuantizationConfig):
awq_marlin_config = AWQMarlinConfig.from_config(
marlin_compatible_config_dict
)
return AWQMarlinMoEMethod(awq_marlin_config, layer.moe_config)
return awq_marlin_config.get_quant_method(layer, prefix)
return None
def apply_vllm_mapper(self, hf_to_vllm_mapper: "WeightsMapper"):