Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Silent crash/segfault when using initial_prompt parameter in model.transcribe() #141

@michaelcrusso

Description

@michaelcrusso

I have some pretty simple code. I've installed with Vulkan support and I'm running on an Nvidia GPU:

segments = model.transcribe(numpy_array_audio, initial_prompt=self.user_prompt, extract_probability=True, token_timestamps=True, max_len=1)
for segment in segments:
    print(segment)
    segmentword = WordObject(word=segment.text, start=segment.t0, end=segment.t1, filename=filename, probability=segment.probability)
    wordresults.append(segmentword)

If I include the initial_prompt parameter, no matter how (I've tried assigning a variable as above, formatting it as an f-string, even just setting initial_prompt = "Test"), the whole program will crash without any exception handling. This happens even if I remove all other params besides initial_prompt. If I run it without initial_prompt, it transcribes without a problem. I don't know if it's an issue with string handling, or my environment? Journalctl shows the following segfault:

Nov 03 00:48:36 michael-ubuntu kernel: python[311683]: segfault at 0 ip 00007fdd904df9b4 sp 00007fdd049b8e98 error 6 in libstdc++-a68762c8.so.6.0.33[df9b4,7fdd9049d000+148000] likely on CPU 8 (core 16, socket 0)

To confirm that this isn't an issue with the Vulkan build specifically, I uninstalled and reinstalled without Vulkan support (no GPU detected, so it should fall back to CPU), and it still caused an instant crash.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions