fixed Llama 2 to 3.2 NBs

[d-kleine]

• fixed RoPE formula (-2)
• updated requirements (safetensors missing, ipywidgets for HF progress bar)
• removed redundant code lines
• removed duplicated imports
• fixed local_dir for llama 3.1 and 3.2 files

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[d-kleine]

About RoPe, I cannot follow the code implementation compared to the formula. This also applies to converting-gpt-to-llama2.ipynb where I think it would be beneficial to add the RoPE formula too.

[rasbt]

Thanks for these refinements!

Regarding

About RoPe, I cannot follow the code implementation compared to the formula. This also applies to converting-gpt-to-llama2.ipynb where I think it would be beneficial to add the RoPE formula too.

Yeah, the formula is only shown to highlight the "theta" and "base" aspects but the one formula is not enough to follow the implementation of the complete RoPE method. I was originally thinking of explaining RoPE here, but this would be better for a separate article because it's a whole topic (or even paper) in itself.

[d-kleine]

Alright, thanks!