Has anyone been able to make word timestamp work? And (UserWarning: Failed to launch Triton kernels..)

[AIhasArrived]

Hello,
I tried to search for similar issues and did not find at first glance,
I am tried this command:
whisper "test2.mp3" --language ... --model large --device cuda -o "test2TranslatedWORDS2" --task transcribe --word_timestamps True --append_punctuations True

But I am not getting timestamps for words, instead I am getting normal time stamps (for sentences).
__
And next to it I am getting this messages:
C:\Users\userA\AppData\Local\Programs\Python\Python311\Lib\site-packages\whisper\timing.py:42: UserWarning: Failed to launch ```
Triton kernels, likely due to missing CUDA toolkit; falling back to a slower median kernel implementation...
warnings.warn(
C:\Users\userA\AppData\Local\Programs\Python\Python311\Lib\site-packages\whisper\timing.py:146: UserWarning: Failed to launch Triton kernels, likely due to missing CUDA toolkit; falling back to a slower DTW implementation...
warnings.warn(

[glangford]

But I am not getting timestamps for words, instead I am getting normal time stamps (for sentences).

Check the .json output, you should see word timestamps there. The .srt file will have normal looking timestamps but refined based on the word timing. To see how accurate the word timing is, you can try --highlight_words True.

...UserWarning: Failed to launch Triton kernels...

This discussion may be helpful

• --word_timestamps True lead to Failed to launch Triton kernels warning #1283

[AIhasArrived]

@glangford thanks for the answer, I have other questions:

1. Do you know about the DLAS library https://github.com/152334H/DL-Art-School (voice clonage)? How do I make all my GPU be used to make the training FASTER there? Any idea?
2. Do you know of other, better, clone voices githubs?

[glangford]

Do you know about the DLAS library

I am not familiar with it.

[dgoryeo]

Does the UserWarning: Failed to launch Triton kernels.. actually impact the performance or output of the transcription? Thanks.