Doubts on the Packages Downloaded for Training and Fine Tuning

[rahulPrakash02]

Hi, @shaheennabi !

First off, hats off to you for your work, I'm trying to train a model using Llama 3.2 instruction tuning for automating code reviews, so this project has been really helpful to me.

But I have a doubt in the first few steps:
I can see that you import !pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118 and !pip install xformers[torch2]. However, immediately in the next few cells we see that you are uninstalling and reinstalling them again, what is the significance of this?

Furthermore, how is torchvision and torchaudio used when we don't deal with image and sound data (from what i can tell anyway)

File for reference: https://github.com/shaheennabi/Production-Ready-Instruction-Finetuning-of-Meta-Llama-3.2-3B-Instruct-Project/blob/main/notebooks/Instruct_Tuning_Llama3_2_3B_Instruct.ipynb

[shaheennabi]

Thanks for reaching out!

The reason I was installing and reinstalling the packages is that while ensuring Unsloth worked perfectly for this use case, I encountered several compatibility issues. Initially, I tried the straightforward installation approach you mentioned, but it didn’t resolve the problems. I also referred to the documentation and attempted other methods, but I still faced challenges.

After experimenting with various combinations, I found that the sequence of installation and reinstallation resolved the issues for me. Specifically, it helped ensure all dependencies were aligned correctly, particularly for Unsloth. If you have a different approach that works for you, feel free to use it! I'd love to hear about it as well.

Regarding the installation of torchvision and torchaudio: these packages are included because xformers and other dependencies have some overlapping requirements. While we aren't directly dealing with image or audio data in this project, ensuring these dependencies are satisfied helps avoid conflicts during runtime.

Thank you once again for your feedback and for taking the time to dive into this project. It really means a lot to me! Let me know if you run into any other issues or have suggestions for improvements etc.

[rahulPrakash02]

Thanks for the quick response, Shaheen! I'll let you know if I have any questions. Regards, Rahul.

…

On Wed, Jan 15, 2025 at 1:10 PM Shaheen Nabi ***@***.***> wrote: Thanks for reaching out! The reason I was installing and reinstalling the packages is that while ensuring Unsloth worked perfectly for this use case, I encountered several compatibility issues. Initially, I tried the straightforward installation approach you mentioned, but it didn’t resolve the problems. I also referred to the documentation and attempted other methods, but I still faced challenges. After experimenting with various combinations, I found that the sequence of installation and reinstallation resolved the issues for me. Specifically, it helped ensure all dependencies were aligned correctly, particularly for Unsloth. If you have a different approach that works for you, feel free to use it! I'd love to hear about it as well. Regarding the installation of torchvision and torchaudio: these packages are included because xformers and other dependencies have some overlapping requirements. While we aren't directly dealing with image or audio data in this project, ensuring these dependencies are satisfied helps avoid conflicts during runtime. Thank you once again for your feedback and for taking the time to dive into this project. It really means a lot to me! Let me know if you run into any other issues or have suggestions for improvements etc. — Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AXDUDKDOSYVWK36KDOOPWHD2KYGGPAVCNFSM6AAAAABVET5NX6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOBTHE4DONY> . You are receiving this because you authored the thread.Message ID: <shaheennabi/Production-Ready-Instruction-Finetuning-of-Meta-Llama-3. ***@***.***>