[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:
Samuel Kim
2026-09-25 02:54:11 +00:00
committed by GitHub
co-authored by mgoin
parent 6936e77ba4
commit f55419c43c
16 changed files with 221 additions and 57 deletions
+2 -2
View File
@@ -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()