【寒武纪硬件产品型号】：MLU370

【使用操作系统】：ubuntu

【使用驱动版本】：v5.10.31

【出错信息】：

[2025-9-4 17:9:34] [CNNL] [Warning]:When calculating multiplication of complex_float data, it is required to use [cnnlGetOpTensorWorkspaceSize_v2] to apply for workspace.

Traceback (most recent call last):

  File "/data/user/RS_IflySFUCoVideo2.0-AudioProcess/DPCodec_mlu/train_dpcodec.py", line 338, in <module>

    trainer.start_train()

  File "/data/user/RS_IflySFUCoVideo2.0-AudioProcess/DPCodec_mlu/train_dpcodec.py", line 323, in start_train

    self.__train()

  File "/data/user/RS_IflySFUCoVideo2.0-AudioProcess/DPCodec_mlu/train_dpcodec.py", line 185, in __train

    self.train_step_apollo(item)

  File "/data/user/RS_IflySFUCoVideo2.0-AudioProcess/DPCodec_mlu/train_dpcodec.py", line 214, in train_step_apollo

    loss_g.backward()

  File "/torch/venv3/pytorch/lib/python3.10/site-packages/torch/_tensor.py", line 495, in backward

    torch.autograd.backward(

  File "/torch/venv3/pytorch/lib/python3.10/site-packages/torch/autograd/__init__.py", line 251, in backward

    Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass

RuntimeError: Can't get a promote dtype when common dtype is ComplexFloat