Fine-tune my Coding-LLM w/ PEFT LoRA Quantization - PART 2

[@javiergimenezmoya86]

Gold. Humanity progres thanks wise people share knoledge and reflects its motivation.

[@damujen]

Hello! I've been thoroughly enjoying your channel and am truly impressed by the innovative and advanced nature of your content. Your ideas are remarkable and it's evident that you put great thought and effort into each video. I do find some of the concepts a bit challenging to fully grasp at times. To help viewers like myself, could you possibly suggest specific areas of study or recommend any courses that would provide us with the foundational knowledge needed to better understand and make the most of your videos? Your channel is an invaluable resource, and I'm keen to deepen my understanding to fully appreciate the insights you share. Thanks for your incredible work!

[@flethacker]

Is there a part 3 showing how to use the model to get code generation in an IDE or with local llama?

[@skanderbegvictor6487]

Yes that would be awesome

[@vivekpadman5248]

This looks gr8, thanks a lot for the code and walkthrough. Can you highlight over the exact structure of the dataset to be used or can any form of dataset can be used to finetune on any model? Also how do we store the qlora weights only or do we have to mix both the models somehow before uploading to the hub?

[@medoeldin]

Thank you for an excellent well structured video. Could you please share your thoughts on the minimum amount of code data that's necessary to get good results fine-tuning Starcoder? You mentioned 3 cloning 3 git hubs wbut was quite sure how much data this entails. Thank you so much.

[@karndeepsingh]

Thanks for the tutorial. I would like to know, how is the instruction dataset prepared for codeLLM?

[@DipeshPaul-x5k]

after training how do i load model, every time i am getting ''size mismatch for base_model' error

[@yashtomar9418]

Hi, I wanted to finetune Star Coder so that it can generate unit test code for a product code, is it possible to achieve that using Star Coder if I give it enough training data of product and unit test code pairs.

[@godspetmonkey]

Tried reaching out to you yesterday via the channel. Let me know if you got the email. MB