[V0 deprecation] Remove QKVCrossParallelLinear implementation (#26475)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
@@ -16,7 +16,6 @@ from compressed_tensors.utils import is_match
|
||||
from vllm.model_executor.layers.linear import (
|
||||
WEIGHT_LOADER_V2_SUPPORTED,
|
||||
LinearMethodBase,
|
||||
QKVCrossParallelLinear,
|
||||
)
|
||||
from vllm.model_executor.layers.quantization.compressed_tensors.compressed_tensors import ( # noqa: E501
|
||||
CompressedTensorsScheme,
|
||||
@@ -89,10 +88,7 @@ class CompressedTensorsLinearTransformMethod(LinearMethodBase):
|
||||
# hack around this by getting weight loader v1 so ULM can load correctly
|
||||
quant_method_name = self.quant_method.__class__.__name__
|
||||
if quant_method_name not in WEIGHT_LOADER_V2_SUPPORTED:
|
||||
if isinstance(layer, QKVCrossParallelLinear):
|
||||
weight_loader_v1 = layer.weight_loader_v1
|
||||
else:
|
||||
weight_loader_v1 = layer.weight_loader
|
||||
weight_loader_v1 = layer.weight_loader
|
||||
extra_weight_attrs["weight_loader"] = weight_loader_v1
|
||||
|
||||
self.quant_method.create_weights(
|
||||
|
||||
Reference in New Issue
Block a user