【TST在windows10上运行错误】RuntimeError: CUDA out of memory.

[bettermorn]

系统环境及依赖包

os : Windows-10-10.0.19042-SP0
python : 3.8.12
tsai : 0.2.24
fastai : 2.5.3
fastcore : 1.3.27
torch : 1.10.1
n_cpus : 8
device : cuda (NVIDIA GeForce MX150)

使用TST模型训练时会出现以下错误

RuntimeError: CUDA out of memory. Tried to allocate 764.00 MiB (GPU 0; 2.00 GiB total capacity; 841.93 MiB already allocated; 277.44 MiB free; 854.00 MiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

原因及解决方法如下

• 参考链接 https://giters.com/CorentinJ/Real-Time-Voice-Cloning/issues/913，方法如下

1. 减小模型训练的batch_size
2. kill 占用GPU内存的进程
3. 增加语句，清除缓存

• 设置 max_split_size_mb
  设置环境变量PYTORCH_CUDA_ALLOC_CONF为max_split_size_mb:128