Distil-Whisper models

[Purfview]

Distil-Whisper models supported since r172.1.

distil-large-v3
distil-large-v2
distil-medium.en
distil-small.en

Distil models for Faster-Whisper require ~twice less memory and they are roughly ~twice faster, when transcription accuracy is somewhere between an original model and a next smaller one.

Note1: These are English only models.
Note2: Just use turbo model, it's better than distill and it's multilingual.

[ghost]

ayyooo whats ur thought rn should we switch to them

[Purfview]

If you don't have hardware issues running the original models then there is no reason to use distill models, as they are less accurate.

[pto2k]

Hi, besides the official/distilled Whisper models, can other fine-tune models be used?
For example: https://huggingface.co/BELLE-2/Belle-whisper-large-v3-zh
Thank you!

[Purfview]

Yes, finetuned models can be used, but your linked model is not faster-whisper model.

[BanQuanChung]

Could you please make a distil-large-v2 model for the Vietnamese langugage?
Thanks.

[Purfview]

Just use turbo model, it's better than distill and it's multilingual.