[BUG] App crashes and can't transribe

[LulaSvob]

Description 🐛

Please provide a clear and concise description of the bug. What went wrong?

Steps to Reproduce 🔍

Steps to reproduce the behavior:

1. Start the project via docker with this line: docker run -d --name scraibe-webui -p 7860:7860 --gpus 'all' -v "$(Get-Location)/data:/data" hadr0n/scraibe-webui
2. Select whisper model "large-v3"
3. Select "keep model alive?"
4. Click apply settings
5. Add a m4a file via the user interface and click transcribe

Expected Behavior 🤔

Should be able to finish the transcription.

Actual Behavior 😕

The browser crashes with "Error code: Crashpad_HandlerDidNotRespond" and then in Docker I get the below errors.

Screenshots or Logs 📸

INFO:speechbrain.utils.quirks:Applied quirks (see `speechbrain.utils.quirks`): [disable_jit_profiling, allow_tf32]
INFO:speechbrain.utils.quirks:Excluded quirks specified by the `SB_DISABLE_QUIRKS` environment (comma-separated list): []

<frozen runpy>:128: RuntimeWarning: 'scraibe_webui.cli' found in sys.modules after import of package 'scraibe_webui', but prior to execution of 'scraibe_webui.cli'; this may result in unpredictable behaviour

INFO:httpx:HTTP Request: GET [https://api.gradio.app/gradio-messaging/en⁠](https://api.gradio.app/gradio-messaging/en) "HTTP/1.1 200 OK"

INFO:httpx:HTTP Request: GET [http://localhost:7860/gradio_api/startup-events⁠](http://localhost:7860/gradio_api/startup-events) "HTTP/1.1 200 OK"

INFO:httpx:HTTP Request: HEAD [http://localhost:7860/⁠](http://localhost:7860/) "HTTP/1.1 200 OK"

INFO:httpx:HTTP Request: GET [https://api.gradio.app/pkg-version⁠](https://api.gradio.app/pkg-version) "HTTP/1.1 200 OK"

/opt/conda/lib/python3.11/site-packages/whisper/__init__.py:150: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See [https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models⁠](https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models) for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.

  checkpoint = torch.load(fp, map_location=device)

/opt/conda/lib/python3.11/site-packages/pyannote/audio/utils/reproducibility.py:74: ReproducibilityWarning: TensorFloat-32 (TF32) has been disabled as it might lead to reproducibility issues and lower accuracy.

It can be re-enabled by calling

   >>> import torch

   >>> torch.backends.cuda.matmul.allow_tf32 = True

   >>> torch.backends.cudnn.allow_tf32 = True

See [https://github.com/pyannote/pyannote-audio/issues/1370⁠](https://github.com/pyannote/pyannote-audio/issues/1370) for more details.


  warnings.warn(

/opt/conda/lib/python3.11/site-packages/pyannote/audio/models/blocks/pooling.py:104: UserWarning: std(): degrees of freedom is <= 0. Correction should be strictly less than the reduction factor (input numel divided by output numel). (Triggered internally at ../aten/src/ATen/native/ReduceOps.cpp:1823.)

  std = sequences.std(dim=-1, correction=1)

/opt/conda/lib/python3.11/site-packages/whisper/__init__.py:150: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See [https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models⁠](https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models) for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.

  checkpoint = torch.load(fp, map_location=device)

/opt/conda/lib/python3.11/site-packages/pyannote/audio/models/blocks/pooling.py:104: UserWarning: std(): degrees of freedom is <= 0. Correction should be strictly less than the reduction factor (input numel divided by output numel). (Triggered internally at ../aten/src/ATen/native/ReduceOps.cpp:1823.)

  std = sequences.std(dim=-1, correction=1)

Traceback (most recent call last):

  File "/opt/conda/lib/python3.11/site-packages/gradio/queueing.py", line 624, in process_events

    response = await route_utils.call_process_api(

               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/opt/conda/lib/python3.11/site-packages/gradio/route_utils.py", line 323, in call_process_api

    output = await app.get_blocks().process_api(

             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/opt/conda/lib/python3.11/site-packages/gradio/blocks.py", line 2015, in process_api

    result = await self.call_function(

             ^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/opt/conda/lib/python3.11/site-packages/gradio/blocks.py", line 1562, in call_function

    prediction = await anyio.to_thread.run_sync(  # type: ignore

                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/opt/conda/lib/python3.11/site-packages/anyio/to_thread.py", line 56, in run_sync

Environment 🖥️

• OS: Windows 11
• Browser: Edge
• Docker Version Version 4.49.0 (208700)
• ScrAIbe-WebUI Version: latest
• Installation Method: Docker

Configuration and Setup ⚙️

• Transcription Model Used (if applicable): Whisper large-v3
• Configuration Details: none - all default

Additional Context 📝

Maybe it's important to note that yesterday I left it to transcribe 3 files by 3 hours each and when I returned at the end of the workday it had crashed and I can't make it work since. Just my thinking ... it may be something about how it stores the files, cache or something, no idea. Deleted the container and re-created it and it still crashes like before but this time it says out of memory, which is more informative:

[JSchmie]

Hi @LulaSvob are you using the image with Nvidia Container Toolkit installed and enabled? Since you are using Windows, what's your way to run Docker (Hyper-V or WSL)?

[LulaSvob]

Hi @LulaSvob are you using the image with Nvidia Container Toolkit installed and enabled? Since you are using Windows, what's your way to run Docker (Hyper-V or WSL)?

Hi @JSchmie, most probably not. I saw this requirement but since the page does not have a Windows 11 in there I don't really know where to install that one. Is it on my WSL? I ruck docker on win 11 with wsl.

[JSchmie]

It should be handled by Docker Desktop, but could you please try running it CPU only?

[LulaSvob]

Same thing. I thought that my ram could have been a blocker and upgraded to 64 ram but am still hitting the out of memory error.
The out of memory error shows up when I use explicit large model. I tried it also with no model selected and got this one: