From d3229431f958a159ac8d9b79969e0068e547d1b0 Mon Sep 17 00:00:00 2001 From: Yongye Zhu Date: Wed, 1 Jul 2026 14:33:51 -0700 Subject: [PATCH] [DSV4] Better MXFP8 quantization kernel (#47229) Signed-off-by: Yongye Zhu --- vllm/model_executor/layers/quantization/utils/mxfp8_utils.py | 1 + 1 file changed, 1 insertion(+) diff --git a/vllm/model_executor/layers/quantization/utils/mxfp8_utils.py b/vllm/model_executor/layers/quantization/utils/mxfp8_utils.py index e6063b4632..d076dce758 100644 --- a/vllm/model_executor/layers/quantization/utils/mxfp8_utils.py +++ b/vllm/model_executor/layers/quantization/utils/mxfp8_utils.py @@ -184,6 +184,7 @@ def _mxfp8_e4m3_quantize_impl( x, is_sf_swizzled_layout=is_sf_swizzled_layout, alignment=alignment if alignment > 0 else 32, + backend="cute-dsl", ) if x_scales.ndim == 1 and x.ndim == 2 and not is_sf_swizzled_layout: x_scales = x_scales.view(x.size(0), -1)