[Kernel] Delegate construction of FusedMoEQuantConfig to FusedMoEMethodBase subclasses (#22537)
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
@@ -11,6 +11,7 @@ from torch.nn.parameter import Parameter
|
||||
from vllm._ipex_ops import ipex_ops as ops
|
||||
from vllm.model_executor.layers.fused_moe import (FusedMoEMethodBase,
|
||||
FusedMoeWeightScaleSupported)
|
||||
from vllm.model_executor.layers.fused_moe.config import FusedMoEQuantConfig
|
||||
from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
|
||||
UnquantizedLinearMethod)
|
||||
from vllm.model_executor.layers.quantization import QuantizationMethods
|
||||
@@ -375,6 +376,10 @@ class XPUFp8MoEMethod(FusedMoEMethodBase):
|
||||
use_prepack=True,
|
||||
)
|
||||
|
||||
def get_fused_moe_quant_config(
|
||||
self, layer: torch.nn.Module) -> Optional[FusedMoEQuantConfig]:
|
||||
return None
|
||||
|
||||
def apply(
|
||||
self,
|
||||
layer: torch.nn.Module,
|
||||
|
||||
Reference in New Issue
Block a user