diff --git a/vllm/model_executor/layers/quantization/utils/mxfp4_utils.py b/vllm/model_executor/layers/quantization/utils/mxfp4_utils.py index 40b26528bc..21c8aba1d5 100644 --- a/vllm/model_executor/layers/quantization/utils/mxfp4_utils.py +++ b/vllm/model_executor/layers/quantization/utils/mxfp4_utils.py @@ -48,9 +48,16 @@ def _swizzle_mxfp4(quant_tensor, scale, num_warps=8): value_layout = StridedLayout if on_gfx950(): - from triton_kernels.tensor_details.layout import GFX950MXScaleLayout + try: + # triton < 3.6 + from triton_kernels.tensor_details.layout import GFX950MXScaleLayout - scale_layout = GFX950MXScaleLayout + scale_layout = GFX950MXScaleLayout + except ImportError: + # triton >= 3.6 + from triton_kernels.tensor_details.layout import CDNA4MXScaleLayout + + scale_layout = CDNA4MXScaleLayout else: scale_layout = StridedLayout else: