-
Notifications
You must be signed in to change notification settings - Fork 9.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Made convert.py work with LLaMA 3 files distributed by meta #7568
Conversation
It should be noted that the |
😭 |
Hmmm... as long as it doesn't break the legacy convert script let's get this merged in quickly before we do the refactor |
That's not a reason to avoid doing a proper review. This looks useful, but there are some parts of it I'd like to review in more detail. Regarding the order of the merges, I think #7430 should be merged first, because it will be much easier to test this with the new file names and locations. (because things will break, especially since |
@Manaball123 Do not lose hope. Programming isn't easy, but that's what makes it fun 😉. @compilade If this is merged, get this merged this first. This should be merged before PR #7430. I'm going to be awhile. I did crack the vocab though. 🥲 My only issue with this PR is that the vocab wasn't properly handled. Other than that, I think it's a good start. It just needs some guidance. |
You could probably merge |
Heads up that #7430 has merged, so will need to refactor this pr |
4f64f7e
to
5921b8f
Compare
uh oh i think i messed something up :( |
To fix this, check your |
Sorry for the inconvenience, made new pr here |
The fix is overall pretty hacky and not very clean, but it does work and doesn't have any pre vocab issues afaik
Improvements such as not using another script can be made, and detection for llama 3 is basically nonexistent, but I guess if someone just wants to get stuff working then this could be applicable
Simple math question test can be seen here(cpu inference, i messed up the cuda build):