关于intel显卡训练失败, 成功生成索引, 但weight文件夹下找不到pth

问题描述:
使用intel Arc 770显卡进行训练, 显示全流程结束, 成功构建索引.
logs文件夹能找到add开头的index索引, 但是weight文件夹下没有pth文件.
命令行窗口过程报错如下:

f0fail-14-C:\software\RVC1006AMD_Intel1/logs/pyxpu/1_16k_wavs/0_124.wav-Traceback (most recent call last):
  File "infer/modules/train/extract/extract_f0_rmvpe.py", line 89, in go
    featur_pit = self.compute_f0(inp_path, f0_method)
  File "infer/modules/train/extract/extract_f0_rmvpe.py", line 52, in compute_f0
    self.model_rmvpe = RMVPE(
  File "C:\software\RVC1006AMD_Intel1\infer\lib\rmvpe.py", line 503, in __init__
    self.mel_extractor = MelSpectrogram(
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 1173, in to
    return self._apply(convert)
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 853, in _apply
    self._buffers[key] = fn(buf)
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 1159, in convert
    return t.to(
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\cuda\__init__.py", line 284, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\utils\weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.
  warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.")
DEBUG:infer.lib.infer_pack.models:gin_channels: 256, self.spk_embed_dim: 109
Process Process-1:
Traceback (most recent call last):
  File "C:\ProgramData\miniconda3\envs\tor\lib\multiprocessing\process.py", line 315, in _bootstrap
    self.run()
  File "C:\ProgramData\miniconda3\envs\tor\lib\multiprocessing\process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "C:\software\RVC1006AMD_Intel1\infer\modules\train\train.py", line 184, in run
    net_g = net_g.cuda(rank)
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 915, in cuda
    return self._apply(lambda t: t.cuda(device))
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 779, in _apply
    module._apply(fn)
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 779, in _apply
    module._apply(fn)
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 804, in _apply
    param_applied = fn(param)
  File "C:\ProgramData\miniconda3\envs\tor\lib\site-packages\torch\nn\modules\module.py", line 915, in <lambda>
    return self._apply(lambda t: t.cuda(device))
RuntimeError: Invalid device, must be cuda device

系统配置
处理器 13th Gen Intel(R) Core(TM) i7-13700K 3.40 GHz
显卡 Intel(R) Arc(TM) A770 Graphics
操作系统 Windows 11 专业版 23H2
rvc代码版本 2.2.231006

安装过程

1. intel显卡驱动已升级到最新32.0.101.6078, 版本: 2024/9/13
2. 已安装oneAPI, 并在oneAPI环境中运行训练,
3. visual studio build tools已添加到系统变量环境

conda create -n tor python==3.8.0
conda activate tor
python -m pip install pip==22.3.1
pip cache purge
conda install pkg-config libuv
pip install torch==2.3.1+cxx11.abi torchvision==0.18.1+cxx11.abi torchaudio==2.3.1+cxx11.abi intel-extension-for-pytorch==2.3.110+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/cn/
pip install -r requirements-dml.txt
pip install --force-reinstall -v "av==11.0.0"
pip install gradio==3.48.0
conda install -c anaconda libjpeg-turbo libpng
pip install transformers

我的疑惑
看到报错是关于cuda, 而intel显卡使用的api是xpu.
找到代码里有关于替换cuda为xpu的ipex模块, 所以应当能够支持intel显卡进行训练, 理想情况下所有cuda均被替换为xpu
但是为什么报错的内容, 提示仅支持cuda设备?
我想弄明白, 究竟是intel显卡设备不支持训练, 还是代码没有将所有cuda替换干净? 亦或是我安装的torch版本有问题?

Start by reproducing the training command in the reported Intel XPU environment and inspect infer/modules/train/extract/extract_f0_rmvpe.py, infer/lib/rmvpe.py, and infer/modules/train/train.py at the traceback locations. Trace the device selection through feature extraction and training; done means the run completes without CUDA-device errors and produces the expected pth file in the weight folder.