[Kernel][Quantization] add w4a8 support for marlin kernel (#24722)

Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
Signed-off-by: Michael Goin <mgoin64@gmail.com>
Signed-off-by: Jinzhen Lin <linjinzhen@hotmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Michael Goin <mgoin@redhat.com>
This commit is contained in:
Jinzhen Lin
2025-11-29 07:19:33 -08:00
committed by GitHub
co-authored by Michael Goin Michael Goin
parent fa59fe417f
commit 1656ad3704
46 changed files with 4370 additions and 2239 deletions
@@ -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, AWQMoEMethod
from .awq_marlin import AWQMarlinConfig, AWQMarlinMoEMethod
from .moe_wna16 import MoeWNA16Config
from .utils.marlin_utils import check_moe_marlin_supports_layer
@@ -136,7 +136,7 @@ class AWQConfig(QuantizationConfig):
awq_marlin_config = AWQMarlinConfig.from_config(
marlin_compatible_config_dict
)
return AWQMoEMethod(awq_marlin_config, layer.moe_config)
return AWQMarlinMoEMethod(awq_marlin_config, layer.moe_config)
return None
def apply_vllm_mapper(self, hf_to_vllm_mapper: "WeightsMapper"):