[INC] fix w4a4 model (#50807)
Signed-off-by: mayuyuace <qiming1.zhang@intel.com>
This commit is contained in:
@@ -496,7 +496,7 @@ def test_qwen3_1p7b_mxfp4_autoround_uses_mxfp4_linear_scheme(
|
|||||||
monkeypatch.setattr(
|
monkeypatch.setattr(
|
||||||
"vllm.model_executor.layers.quantization.inc.schemes."
|
"vllm.model_executor.layers.quantization.inc.schemes."
|
||||||
"inc_mxfp4_linear.init_mxfp4_linear_kernel",
|
"inc_mxfp4_linear.init_mxfp4_linear_kernel",
|
||||||
lambda: DummyKernel(),
|
lambda **kwargs: DummyKernel(),
|
||||||
)
|
)
|
||||||
|
|
||||||
from vllm.model_executor.layers.quantization.inc.schemes.inc_mxfp4_linear import ( # noqa: E501
|
from vllm.model_executor.layers.quantization.inc.schemes.inc_mxfp4_linear import ( # noqa: E501
|
||||||
@@ -614,7 +614,7 @@ def test_inc_mxfp4_linear_method_registers_and_processes_weights(
|
|||||||
monkeypatch.setattr(
|
monkeypatch.setattr(
|
||||||
"vllm.model_executor.layers.quantization.inc.schemes."
|
"vllm.model_executor.layers.quantization.inc.schemes."
|
||||||
"inc_mxfp4_linear.init_mxfp4_linear_kernel",
|
"inc_mxfp4_linear.init_mxfp4_linear_kernel",
|
||||||
lambda: DummyKernel(),
|
lambda **kwargs: DummyKernel(),
|
||||||
)
|
)
|
||||||
monkeypatch.setattr(
|
monkeypatch.setattr(
|
||||||
"vllm.model_executor.parameter.get_tensor_model_parallel_rank",
|
"vllm.model_executor.parameter.get_tensor_model_parallel_rank",
|
||||||
|
|||||||
@@ -7,6 +7,7 @@ import torch
|
|||||||
from torch.nn.parameter import Parameter
|
from torch.nn.parameter import Parameter
|
||||||
|
|
||||||
from vllm.model_executor.kernels.linear import init_mxfp4_linear_kernel
|
from vllm.model_executor.kernels.linear import init_mxfp4_linear_kernel
|
||||||
|
from vllm.model_executor.layers.quantization.utils.quant_utils import kMxfp4Dynamic
|
||||||
from vllm.model_executor.parameter import (
|
from vllm.model_executor.parameter import (
|
||||||
GroupQuantScaleParameter,
|
GroupQuantScaleParameter,
|
||||||
ModelWeightParameter,
|
ModelWeightParameter,
|
||||||
@@ -29,7 +30,7 @@ class INCMxfp4LinearMethod(INCLinearScheme):
|
|||||||
|
|
||||||
def __init__(self, layer_config: "INCLayerConfig") -> None:
|
def __init__(self, layer_config: "INCLayerConfig") -> None:
|
||||||
self.group_size = layer_config.group_size or 32
|
self.group_size = layer_config.group_size or 32
|
||||||
self.kernel = init_mxfp4_linear_kernel()
|
self.kernel = init_mxfp4_linear_kernel(activation_quant_key=kMxfp4Dynamic)
|
||||||
|
|
||||||
@classmethod
|
@classmethod
|
||||||
def get_min_capability(cls) -> int:
|
def get_min_capability(cls) -> int:
|
||||||
|
|||||||
Reference in New Issue
Block a user