Purfview
Purfview
>when I comment out Pipeline.from_pretrained row, the script runs successfully Did you tried commenting out faster-whisper instead?
>model = WhisperModel("large-v3", device=Device, device_index=[0,1], compute_type="int8_float16") Try to remove "device_index=[0,1]"
Not whole, just the reference view, turn off the reference view till the fix.
>It doesn't even support the dark mode layout, it seems. It didn't support it before too, IMDb Scout Mod was making it dark.
Fixed in v25.0
Ha, the unhappy customers... Looks like [my comment](https://github.com/SYSTRAN/faster-whisper/pull/856#issuecomment-2160472018) was prophetic. 😜 I still think that merging that PR was not wise decision, some things there are suboptimal and breaking for...
@aligokalppeker That would mean that he got privileges to manage this repo, that's good thing that more people joined management team. Btw, I didn't followed much what is going on...
>Purfview also wrote that the PR contains multiple unrelated stuff. That was before PR was cleaned up. And there is PR https://github.com/SYSTRAN/faster-whisper/pull/936 to remove redundant VAD and maybe torch dependency...
>How do I ensure that it is actually running with CUDA GPU acceleration? model = WhisperModel(model_size, device="cuda") > Inference time is identical to CPU execution Did you tested long audio?...
>No, tested on multiple small audios of 2-3 secs loaded in a folder. Ah, you need to test much longer audios. For 2-3 secs the model loading/conversion & ect can...