Please Release Fine-Tuning Scripts

Question

Please Release Fine-Tuning Scripts

ohmguru opened this issue 10 months ago · comments

ohmguru commented 10 months ago

Please release fine-tuning scripts so we can adapt this for medium and large whisper models.

Alexandre Cassagne commented 8 months ago

+1

Harper Grieve · Answer 1 · Tue Aug 01 2023 00:18:55 GMT+0800 (China Standard Time)

Please! I would love to fine tune and contribute the larger models!

Matthew Campbell · Answer 2 · Tue Aug 01 2023 19:00:31 GMT+0800 (China Standard Time)

Yeah I'm looking to port this to other human languages. Would be nice to know how it's done. it seems like the project is on ice right now?

Akash Mahajan · Answer 3 · Thu Aug 17 2023 01:36:45 GMT+0800 (China Standard Time)

Hi folks - sorry about the delay, i've been on a break for a bit with some personal life updates. Ack - and will keep you posted. Appreciate the patience/interest!

qhkm · Answer 4 · Mon Aug 21 2023 16:17:01 GMT+0800 (China Standard Time)

Would love to contribute! Waiting for the fine tuning scripts.

Akash Mahajan · Answer 5 · Wed Nov 01 2023 15:11:13 GMT+0800 (China Standard Time)

Hi everyone,

Firstly, thanks a lot for your interest in this project!

As you may have noticed, releases I'd initially planned have been on a pause over the last couple months, particularly the request for release of finetuning code to reproduce results shared in the repository.

Due to discussions on professional constraints with my employer, I've had to be conservative and refrain from making any major releases. My apologies as I didn't anticipate an issue here given that I personally found finetuning relatively simple to implement, and cheap to run on a consumer GPU. If you're interested in trying on your end, feel free to dig around on the repo a bit to get an idea how to start. However unfortunately until the issue on my end is resolved I recommend assuming an indefinite pause on further major releases from me.

I can imagine this is not news you'd have liked to hear, but I wanted to be transparent here and trust that you'd be able to understand. Much appreciated! 🙏

Best,
Akash

Jordi Bruin · Answer 6 · Wed Nov 01 2023 22:01:15 GMT+0800 (China Standard Time)

Thanks for the reply, totally get it!

Rahul Somani · Answer 7 · Wed Nov 08 2023 05:58:00 GMT+0800 (China Standard Time)

Thanks for the thoughtful response and transparency Akash, that's totally understandable!

Vaibhav Srivastav · Answer 8 · Tue Nov 28 2023 18:55:45 GMT+0800 (China Standard Time)

Hey @akashmjn - I'm VB, I lead the advocacy effort for open source audio at Hugging Face. It's sad to see that you've had to cut down on major releases because of professional reasons. If you're game then we'd love to help scale your experiments to large-v3 checkpoint. I think it'd be a huge win for the community.

Feel free to DM me at reach_vb and we can work something out! Hoping we can work something out that works for your constraints and the community too! 🤗

Open source for the win!

Mokshith Voodarla · Answer 9 · Sun Dec 03 2023 02:09:46 GMT+0800 (China Standard Time)

This project is insanely cool. Just thought I'd leave a comment!

Akash Mahajan · Answer 10 · Wed Dec 13 2023 04:34:32 GMT+0800 (China Standard Time)

Hey @akashmjn - I'm VB, I lead the advocacy effort for open source audio at Hugging Face. It's sad to see that you've had to cut down on major releases because of professional reasons. If you're game then we'd love to help scale your experiments to large-v3 checkpoint. I think it'd be a huge win for the community.

Feel free to DM me at reach_vb and we can work something out! Hoping we can work something out that works for your constraints and the community too! 🤗

Open source for the win!

Thanks for the note @Vaibhavs10, and excellent to see the interest in scaling this up from HuggingFace! Dropped you a DM and let's see what we can work out.