[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
@@ -266,7 +266,7 @@ class AutoRoundConfig(QuantizationConfig):
from vllm.model_executor.layers.quantization.awq_marlin import (
AWQMarlinConfig,
AWQMarlinLinearMethod,
AWQMoEMethod,
AWQMarlinMoEMethod,
)
quant_args_marlin = AWQMarlinConfig(
@@ -291,7 +291,7 @@ class AutoRoundConfig(QuantizationConfig):
if isinstance(layer, FusedMoE):
if use_marlin:
return AWQMoEMethod(quant_args_marlin, layer.moe_config)
return AWQMarlinMoEMethod(quant_args_marlin, layer.moe)
from vllm.model_executor.layers.quantization.moe_wna16 import MoeWNA16Config
config = {