From 6326222d68a6fda9b7aab44ce3b8f0eb9ac75e21 Mon Sep 17 00:00:00 2001 From: biondizzle Date: Mon, 18 May 2026 20:40:48 +0000 Subject: [PATCH] Fix: add abstract create_weights to CuTeDSLNvfp4LinearMethod --- vllm/cutedsl_quant_method.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/vllm/cutedsl_quant_method.py b/vllm/cutedsl_quant_method.py index ce6aaa1d..c5d91873 100644 --- a/vllm/cutedsl_quant_method.py +++ b/vllm/cutedsl_quant_method.py @@ -131,5 +131,10 @@ class CuTeDSLNvfp4Method(LinearMethodBase): class CuTeDSLNvfp4LinearMethod(LinearMethodBase): """Forward path: BF16 input → CuTeDSL NVFP4 GEMM → BF16 output.""" + def create_weights(self, layer, input_size_per_partition, + output_partition_sizes, input_size, output_size, + params_dtype, **extra_weight_attrs): + pass + def apply(self, layer, x: torch.Tensor, bias=None) -> torch.Tensor: return layer._cutedsl_runner(x)