Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

insanely fast Whisper error #486

Open
xu2k opened this issue Jan 28, 2025 · 6 comments
Open

insanely fast Whisper error #486

xu2k opened this issue Jan 28, 2025 · 6 comments
Assignees
Labels
bug Something isn't working

Comments

@xu2k
Copy link

xu2k commented Jan 28, 2025

자막\Whisper-WebUI-Portable-Windows\Whisper-WebUI\venv\Lib\site-packages\transformers\models\whisper\generation_whisper.py:573: FutureWarning: The input name inputs is deprecated. Please make sure to use input_features instead.
warnings.warn(
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:00:01You have passed task=transcribe, but also have set forced_decoder_ids to [[1, None], [2, 50359]] which creates a conflict. forced_decoder_ids will be ignored in favor of task=transcribe.
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:00:43Passing a tuple of past_key_values is deprecated and will be removed in Transformers v4.43.0. You should pass an instance of EncoderDecoderCache instead, e.g. past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values).
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:00:49The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's attention_mask to obtain reliable results.
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:03:33From v4.47 onwards, when a model cache is to be returned, generate will return a Cache instance instead by default (as opposed to the legacy tuple of tuples format). If you want to keep returning the legacy format, please set return_legacy_cache=True

이런식으로 무한 로딩이 되는데 어떡하죠??

@xu2k xu2k added the bug Something isn't working label Jan 28, 2025
@jhj0517
Copy link
Owner

jhj0517 commented Jan 29, 2025

안녕하세요.

버그 재현을 위해 더 상세한 로그가 있으면 좋겠는데요, 올려주신게 터미널에 보이는 전부 인가요?

Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:03:33 From v4.47 onwards,

로그의 시간대로 짐작 가는 바로는 모델 로딩에 어려움을 겪고 있는 것 같아요. insanely_fast_whisper 의 경우 large-v2 모델을 사용했을 때 피크 VRAM 이 10.5GB 까지 오르는 것을 확인했습니다.

혹시 사용하시는 GPU의 VRAM 이 어떻게 되시나요?

@xu2k
Copy link
Author

xu2k commented Jan 29, 2025

gtx 1660 ti에 6gb입니다

미디엄 사용하고 있고요

@xu2k
Copy link
Author

xu2k commented Jan 29, 2025

방금 다시 한번 시도해보니

C:\Users\sgwli\OneDrive\바탕 화면\ai
자막\Whisper-WebUI-Portable-Windows\Whisper-WebUI\venv\Lib\site-packages\transformers\models\whisper\generation_whisper.
py:573: FutureWarning: The input name inputs is deprecated. Please make sure to use input_features instead.
warnings.warn(
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:00:00You have passed task=transcribe, but also have set forced_decoder_ids to [[1, None], [2, 50359]] which creates a conflict. forced_decoder_ids will be ignored in favor of task=transcribe.
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:00:43Passing a tuple of past_key_values is deprecated and will be removed in Transformers v4.43.0. You should pass an instance of EncoderDecoderCache instead, e.g. past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values).
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:00:50The attention mask is not set and cannot be inferred from input because pad token is same as eos token. As a consequence, you may observe unexpected behavior. Please pass your input's attention_mask to obtain reliable results.
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:03:58From v4.47 onwards, when a model cache is to be returned, generate will return a Cache instance instead by default (as opposed to the legacy tuple of tuples format). If you want to keep returning the legacy format, please set return_legacy_cache=True.
Transcribing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0:05:49

이렇게 뜨고 무한으로 진행중입니다 ㅠ

@jhj0517
Copy link
Owner

jhj0517 commented Jan 30, 2025

insanely_fast_whisper 에서 미디엄 모델 사용 시 피크 VRAM 이 3.1 GB 인 것을 확인했지만, 코랩의 T4 GPU 런타임에서 실행한 것이라 정확한 버그 재현이 어렵네요..

문제가 없다면 faster-whisper 를 사용해보시겠어요? faster-whisper에 의하면 large-v2 모델을 사용해도 피크 VRAM 이 4.7GB 밖에 되지 않기 때문에, 아마도 6GB VRAM 의 환경에서도 문제 없이 돌아갈 것입니다.

@xu2k
Copy link
Author

xu2k commented Jan 31, 2025

이미 기본값인 faster-whisper을 사용하고 있기는 한데.. 좀 느린 거 같아서요

그래서 insanely를 사용해보려고 했는데 어쩔 수 없네요

@jhj0517
Copy link
Owner

jhj0517 commented Jan 31, 2025

GPU 가 좀 느린 감이 있다면 코랩에서 무료버전 런타임을 사용하시는게 좋으실 수도 있어요.

"T4 GPU" 런타임 (무료) 을 선택하신 후 셀을 순서대로 실행하시면 됩니다.

파일 용량이 커서 업로드에 시간이 걸린다면 구글 드라이브에 먼저 업로드 후 해당 폴더 경로를 "Input Folder Path" 에 입력하는 방식으로 사용할 수도 있어요.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

2 participants