[INC] fix w4a4 model (#50807)
Signed-off-by: mayuyuace <qiming1.zhang@intel.com>
Qiming Zhang committed
Aug 2, 2026 at 23:28 UTC
2755489a261b15ddcd8ced19f0bdf4ea63aad688
2 files changed
+4
-3
tests/quantization/test_auto_round.py
+2
-2
@@ -496,7 +496,7 @@ def test_qwen3_1p7b_mxfp4_autoround_uses_mxfp4_linear_scheme(
496
monkeypatch.setattr(
497
"vllm.model_executor.layers.quantization.inc.schemes."
498
"inc_mxfp4_linear.init_mxfp4_linear_kernel",
499
- lambda: DummyKernel(),
499
+ lambda **kwargs: DummyKernel(),
500
)
501
502
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(
614
monkeypatch.setattr(
615
"vllm.model_executor.layers.quantization.inc.schemes."
616
"inc_mxfp4_linear.init_mxfp4_linear_kernel",
617
- lambda: DummyKernel(),
617
+ lambda **kwargs: DummyKernel(),
618
)
619
monkeypatch.setattr(
620
"vllm.model_executor.parameter.get_tensor_model_parallel_rank",
vllm/model_executor/layers/quantization/inc/schemes/inc_mxfp4_linear.py
+2
-1
@@ -7,6 +7,7 @@ import torch
7
from torch.nn.parameter import Parameter
8
9
from vllm.model_executor.kernels.linear import init_mxfp4_linear_kernel
10
+from vllm.model_executor.layers.quantization.utils.quant_utils import kMxfp4Dynamic
11
from vllm.model_executor.parameter import (
12
GroupQuantScaleParameter,
13
ModelWeightParameter,
@@ -29,7 +30,7 @@ class INCMxfp4LinearMethod(INCLinearScheme):
30
31
def __init__(self, layer_config: "INCLayerConfig") -> None:
32
self.group_size = layer_config.group_size or 32
32
- self.kernel = init_mxfp4_linear_kernel()
33
+ self.kernel = init_mxfp4_linear_kernel(activation_quant_key=kMxfp4Dynamic)
34
35
@classmethod
36
def get_min_capability(cls) -> int: