Update model.py

[jeremylaratro]

added default True to use_scaled_rope due to issues surrounding this function.

[vekoada]

Addresses #291

[vekoada]

Thanks @jeremylaratro

This addresses a real issue (TypeError) when running example_text_completion.py (Addresses #291). Turns out params.json had use_scaled_rope, but ModelArgs was missing it. The llama model describe command shows that this parameter is part of the intended configuration, and it's a straightforward fix.

I'd say this is good to merge. It would be awesome if you could add a quick comment in the code just to explain what use_scaled_rope does.

[galeselee]

This gets llama3.1 running, but it doesn't use the scaled rope

[jeremylaratro]

As far as I can tell - the actual logic for scaled_rope has not been added yet, causing the code to break when trying to run the model (as @galeselee said). This was causing a lot of headaches (as can be seen online), so I added this quick fix to make the model usable without effecting/removing any parameters.

The other two options seem to be:
a) remove the scaled rope parameters from model.py and params.json entirely
b) implement scaled rope logic

This fix keeps those scaled_rope parameters there for anyone who wishes to implement or utilize scaled_rope [should it be added], while allowing other users to run the model without having to debug and make the changes on their own every time they clone the repo.

@vekoada - I can add a comment summarizing this for the time being. I did make an attempt at implementing scaled_rope, but I have more testing to do as this is not my niche.