From 3c95a87b662fd0b97b9ffe8cad3ec93ac3b48eff Mon Sep 17 00:00:00 2001 From: Brayden Zhong Date: Wed, 24 Jun 2026 15:26:46 -0700 Subject: [PATCH] Fix the CuDNN failure on bmm_fp8 when two libcudart.so exists. (#29201) Co-authored-by: Brayden Zhong --- python/sglang/srt/layers/quantization/fp8_utils.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/layers/quantization/fp8_utils.py b/python/sglang/srt/layers/quantization/fp8_utils.py index 837f3894b..70cff5edf 100755 --- a/python/sglang/srt/layers/quantization/fp8_utils.py +++ b/python/sglang/srt/layers/quantization/fp8_utils.py @@ -270,7 +270,7 @@ if is_blackwell_supported() and is_flashinfer_available(): x_scale.reshape(1), weight_scale.reshape(1), out_dtype, - backend="auto", + backend="cublas", ).view(m, n) @lru_cache(maxsize=1)