diff --git a/vllm/model_executor/warmup/deep_gemm_warmup.py b/vllm/model_executor/warmup/deep_gemm_warmup.py index e0c584df8760b..936f6b1e28ce1 100644 --- a/vllm/model_executor/warmup/deep_gemm_warmup.py +++ b/vllm/model_executor/warmup/deep_gemm_warmup.py @@ -89,7 +89,7 @@ def _extract_data_from_linear_base_module( assert m.quant_method.quant_config is not None w = m.weight - ws = m.weight_scale + ws = m.weight_scale_inv if hasattr(m, "weight_scale_inv") else m.weight_scale quant_block_size = m.quant_method.quant_config.weight_block_size assert isinstance(w, torch.Tensor)