Demucs 报错

[colin4k]

内存爆了，但是我64G 内存。。。
Traceback (most recent call last):
File "D:\Soft\anaconda3\envs\videolingo\lib\site-packages\streamlit\runtime\scriptrunner\exec_code.py", line 88, in exec_func_with_error_handling
result = func()
File "D:\Soft\anaconda3\envs\videolingo\lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 590, in code_to_exec
exec(code, module.dict)
File "D:\workspace\VideoLingo-1.7.0\st.py", line 117, in
main()
File "D:\workspace\VideoLingo-1.7.0\st.py", line 113, in main
text_processing_section()
File "D:\workspace\VideoLingo-1.7.0\st.py", line 30, in text_processing_section
process_text()
File "D:\workspace\VideoLingo-1.7.0\st.py", line 45, in process_text
step2_whisper.transcribe()
File "D:\workspace\VideoLingo-1.7.0\core\step2_whisper.py", line 13, in transcribe
ts(video_file)
File "D:\workspace\VideoLingo-1.7.0\core\all_whisper_methods\whisperX.py", line 126, in transcribe
demucs_main(
File "D:\workspace\VideoLingo-1.7.0\core\all_whisper_methods\demucs_vl.py", line 61, in demucs_main
_, outputs = separator.separate_audio_file(music_file)
File "D:\Soft\anaconda3\envs\videolingo\lib\site-packages\demucs\api.py", line 308, in separate_audio_file
return self.separate_tensor(self._load_audio(file), self.samplerate)
File "D:\Soft\anaconda3\envs\videolingo\lib\site-packages\demucs\api.py", line 271, in separate_tensor
out = apply_model(
File "D:\Soft\anaconda3\envs\videolingo\lib\site-packages\demucs\apply.py", line 216, in apply_model
res = apply_model(sub_model, mix, **kwargs, callback_arg=callback_arg)
File "D:\Soft\anaconda3\envs\videolingo\lib\site-packages\demucs\apply.py", line 253, in apply_model
out += shifted_out[..., max_shift - offset:]
RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 42681172032 bytes.

[colin4k]

重新启动看了下，系统本身日常占用 6G 多内存，启动 Videilingo 后变成 12G 多，开始Demucs 后马上把 60G+内存全部占满，然后报错

[Huanshere]

这很奇怪，这是个很成熟的库，而且用法也很简单，我暂时没有复现过

[colin4k]

这很奇怪，这是个很成熟的库，而且用法也很简单，我暂时没有复现过

我估计是因为处理的对象是一个 2g 多时长 8 个多小时的文件

[Huanshere]

没有测试过这么长的文件，但是在处理的时候是会切分音频的，理论上也不会爆内存（）