[DSV4] Better MXFP8 quantization kernel (#47229)
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Yongye Zhu committed
Jul 1, 2026 at 14:33 UTC
d3229431f958a159ac8d9b79969e0068e547d1b0
1 file changed
+1
vllm/model_executor/layers/quantization/utils/mxfp8_utils.py
+1
@@ -184,6 +184,7 @@ def _mxfp8_e4m3_quantize_impl(
184
x,
185
is_sf_swizzled_layout=is_sf_swizzled_layout,
186
alignment=alignment if alignment > 0 else 32,
187
+ backend="cute-dsl",
188
)
189
if x_scales.ndim == 1 and x.ndim == 2 and not is_sf_swizzled_layout:
190
x_scales = x_scales.view(x.size(0), -1)