[Bugfix][Quantization] Fix MXFP8 startup crash on layers below mm_mxfp8 shape limits (#54223)
Signed-off-by: samuelkim7 <samuelmwkim@gmail.com> Signed-off-by: mgoin <mgoin64@gmail.com> Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -1249,7 +1249,7 @@ def test_inc_mxfp8_linear_scheme_delegates_to_kernel(monkeypatch) -> None:
|
||||
kernel = DummyKernel()
|
||||
monkeypatch.setattr(
|
||||
"vllm.model_executor.layers.quantization.inc.schemes.inc_mxfp8_linear.init_mxfp8_linear_kernel",
|
||||
lambda: kernel,
|
||||
lambda weight_shape: kernel,
|
||||
)
|
||||
monkeypatch.setattr(
|
||||
"vllm.model_executor.layers.quantization.inc.schemes.inc_mxfp8_linear.ModelWeightParameter",
|
||||
@@ -1288,7 +1288,7 @@ def test_inc_mxfp8_linear_scheme_delegates_to_kernel(monkeypatch) -> None:
|
||||
def test_inc_mxfp8_linear_scheme_requires_block_32_input(monkeypatch) -> None:
|
||||
monkeypatch.setattr(
|
||||
"vllm.model_executor.layers.quantization.inc.schemes.inc_mxfp8_linear.init_mxfp8_linear_kernel",
|
||||
lambda: object(),
|
||||
lambda weight_shape: object(),
|
||||
)
|
||||
scheme = INCMxfp8LinearScheme()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user