modelopt.torch.quantization.backends.nvfp4_gemm# nvfp4_gemm(quant_module, input_tensor, bias=None)# GEMM function for fp4 quantization.