Streamlit memory error

[ducminh1712]

After inputting the media path, hitting Enter on the web ui, the memory usage is through the roof. And it crashed. Here is the full error showing in console.

(subsai) D:\Workspace>subsai-webui
[17:46:06] WARNING  The torchaudio backend is switched to 'soundfile'. Note that 'sox_io' is   torch_audio_backend.py:19
                    not supported on Windows.
[17:46:46] WARNING  The torchaudio backend is switched to 'soundfile'. Note that 'sox_io' is   torch_audio_backend.py:19
                    not supported on Windows.
2023-07-11 17:46:47.647 WARNING streamlit.runtime.caching.cache_data_api: No runtime found, using MemoryCacheStorageManager

  You can now view your Streamlit app in your browser.

  Local URL: http://localhost:8501
  Network URL: http://192.168.31.61:8501

Warning: temperature does not have a supported UI
Warning: temperature does not have a supported UI
2023-07-11 17:48:36.401 Uncaught app exception
Traceback (most recent call last):
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\streamlit\components\v1\components.py", line 151, in create_instance
    serialized_json_args = json.dumps(json_args)
                           ^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\json\__init__.py", line 231, in dumps
    return _default_encoder.encode(obj)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\json\encoder.py", line 200, in encode
    chunks = self.iterencode(o, _one_shot=True)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\json\encoder.py", line 258, in iterencode
    return _iterencode(o, 0)
           ^^^^^^^^^^^^^^^^^
MemoryError

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 565, in _run_script
    exec(code, module.__dict__)
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\subsai\webui.py", line 501, in <module>
    run()
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\subsai\webui.py", line 494, in run
    webui()
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\subsai\webui.py", line 470, in webui
    event = st_player(_media_file_base64(file_path), **options, height=500, key="player")
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\streamlit_player\__init__.py", line 84, in st_player
    event = _component_func(
            ^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\streamlit\components\v1\components.py", line 79, in __call__
    return self.create_instance(*args, default=default, key=key, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\streamlit\runtime\metrics_util.py", line 311, in wrapped_func
    result = non_optional_func(*args, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\minh\.conda\envs\subsai\Lib\site-packages\streamlit\components\v1\components.py", line 153, in create_instance
    raise MarshallComponentException(
streamlit.components.v1.components.MarshallComponentException: ('Could not convert component args to JSON', MemoryError())

[abdeladim-s]

@ducminh1712,

From what I can see, the cause of the error is the st_player component!

Are you using a huge media file by any chance ? Can you try with a small media file ?

[ducminh1712]

Yes, it was a big media file. Would the CLI be a workaround for this at least for now? If it is, would be great if a --verbose option could be added to track the progress.

[abdeladim-s]

• Yes I think with the CLI, it will work.

• Not all models provide a verbose option, but if a model did -like the original Whisper- (check the wanted model configs in the docs to see the different options) you can pass it to the CLI as follows:

  subsai ./assets/test1.mp4 --model openai/whisper --model-configs '{"verbose": true}' --format srt

• Another thing that might help in this case is to use the script I provided in the examples section for processing long files using silero VAD: VAD example: process long audio files.

Hope this helps!