[TTS]华为昇腾服务器 910B 错误 ACL error, the error code is : 100000

[lezhizhe]

服务器信息

• 镜像 PyTorch 2.1.0 Python 3.10(ubuntu22.04) CANN 8.0.0
• GPU：910B2x鲲鹏920(64GB) * 1
• CPU：24 vCPU Kunpeng-920

安装代码：

python -m pip install paddlepaddle==3.0.0rc1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
python -m pip install paddle-custom-npu==3.0.0rc1 -i https://www.paddlepaddle.org.cn/packages/stable/npu/
git clone https://github.com/PaddlePaddle/PaddleSpeech.git ./speech
pip install pytest-runner
pip install -e . --use-pep517

执行测试代码TTS错误：

import paddle
from paddlespeech.cli.tts.infer import TTSExecutor
 
print(f"Current device: {paddle.device.get_device()}")

tts = TTSExecutor()

tts(text="今天天气十分不错。", output="output.wav")

错误提示：

Traceback (most recent call last):
  File "/root/speech/tts.py", line 9, in <module>
    tts(text="今天天气十分不错。", output="output.wav")
  File "/root/speech/paddlespeech/cli/utils.py", line 328, in _warpper
    return executor_func(self, *args, **kwargs)
  File "/root/speech/paddlespeech/cli/tts/infer.py", line 696, in __call__
    self._init_from_path(
  File "/root/speech/paddlespeech/cli/tts/infer.py", line 337, in _init_from_path
    self.am_inference = get_am_inference(
  File "/root/speech/paddlespeech/t2s/exps/syn_utils.py", line 405, in get_am_inference
    am = am_class(
  File "/root/speech/paddlespeech/t2s/models/fastspeech2/fastspeech2.py", line 516, in __init__
    self._reset_parameters(
  File "/root/speech/paddlespeech/t2s/models/fastspeech2/fastspeech2.py", line 908, in _reset_parameters
    init_enc_alpha = paddle.to_tensor(init_enc_alpha).reshape([1])
  File "/root/miniconda3/lib/python3.10/site-packages/paddle/tensor/manipulation.py", line 4988, in reshape
    out = _C_ops.reshape(x, new_shape)
OSError: (External)  ACL error, the error code is : 100000.  (at /paddle/backends/npu/kernels/funcs/npu_op_runner.cc:225)

[zxcd]

该模型暂未做针对 NPU 的适配，如果要用可能需要自己 debug 一下。如果对 npu debug了解较少，可以去 paddle 主 repo 下提出进一步 issue。