Add LoRA, QLoRA fine-tuning for HF models

[VictorOdede]

Current onsite LLM class uses full parameter fine-tuning which costly. LoRA fine-tuning will require less memory and prevent overfitting by freezing the pretrained weights.

[cartazio]

yeah, sounds good

[bilal-aamer]

Hey @VictorOdede, I'd like to work on this.

[VictorOdede]

Kindly mention this in the engineering channel on Discord then @mmirman or @cartazio will give you the green light.

Hey @VictorOdede, I'd like to work on this.

[mmirman]

@bilal-aamer We certainly won't stop you from working on it :-) but it'd be good to get on a call with @abhigya-sodani to see what is useful to know before starting it!

[bilal-aamer]

Sure will do! Did some due diligence already, need some pointers from @abhigya-sodani.

[abhigya-sodani]

sure bilal lets get on a call soon

[harivamsi9]

If someone is not actively working on this, I can take up this task

[cartazio]

Have at it!

…

On Mon, Aug 14, 2023 at 9:47 AM Hari Vamsi ***@***.***> wrote: If someone is not actively working on this, I can take up this task — Reply to this email directly, view it on GitHub <#97 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAABBQVSV53J3WQS3P3EZLTXVIT65ANCNFSM6AAAAAA2XRNFKI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[abhigya-sodani]

#7 is also applicable/related to this ticket

[berkay3500]

Is there any team working on this issue?

[cartazio]

if youre confident you can bang something together, have at it! this one of the more complex open tickets, so be honest with yourself! that said, worst case you learn something!