Spaces:
Sleeping
Apply for community grant: Personal project
Hi @reach-vb @sanchit-gandhi ,
Thank you both for organising this great event!
I have fine-tuned several Whisper models (small&medium&large) on French and made this demo. In this space I also added an option to choose different models to inference with (using the dropdown "Whisper Model"), in order to compare different fine-tuned models and pretrained models on the same audios.
Right now the inference is taking so long on 2 CPUs, I'm wondering if you could grant GPU to this space. I will upload my large checkpoint when it's finished and wish to test it with this space.
Thanks in advance :)
For the moment the best fine-tuned whisper-medium model got the following WERs on MCV 11.0
- 10.67% greedy
- 10.07% beam
Compare to WERs reported in the paper on MCV 9.0
- 16.0% with whisper-medium
- 13.9% with whisper-large-v2
Hi
@bofenghuang
- This is fantastic! Your model looks fantastic, thank you for putting in time and effort into fine-tuning the model. We've decided to assign you a T4 instance for the demo. This grant will run till 23rd December 2022. After which we'll continue the grant if there's enough traction on the space.
Can't wait to see what language/ model you tackle next. Thanks again!
Don't forget to tweet/ LinkedIn your space, I'm sure the community would love to try it. Tag @huggingface and @lambdaapi when you do! 🤗