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
Add support for Llama 3 (and Llama-2-70b-hf) #549
Conversation
llama-2-70b-hf and Llama 3 models all have n_key_value_heads set so they'll use Grouped-Query Attention.
Looks like the docs fail to build because "Repo model meta-llama/Llama-2-7b-hf is gated. You must be authenticated to access it." Might be the same as #548. |
Thanks so much for the PR! Yeah our solution to the docs thing is to hard
code the config as a dict. It's hacky, but it's not like the config is ever
going to change. Can you do that? Should be easy enough to copy what the
existing llama models do
…On Sat, 20 Apr 2024, 8:32 pm Joel Burget, ***@***.***> wrote:
Looks like the docs fail to build because "Repo model
meta-llama/Llama-2-7b-hf is gated. You must be authenticated to access it."
Might be the same as #548
<#548>.
—
Reply to this email directly, view it on GitHub
<#549 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ASRPNKNNW3VLI5Y7JAPVKCTY6K7FDAVCNFSM6AAAAABGQV45CGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANRXG43DGMJQGQ>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
See discussion on TransformerLensOrg#549 for why.
Sure thing! Hopefully what I have now looks okay. Note that the docs are failing to build but I think it's because of #548:
|
Note in case someone needs to do this in the future: Since the Llama 3 configs are exactly the same as the other Llama models (
Then, on a computer signed in to a HF account with access:
|
Yeah, that config is definitely a bit unruly. Revising it to find ways to eliminate shared code, or finding other ways to make it more manageable is a worthwhile undertaking. The docs issue is resolved. As long as everything is still passing with the recent changes, I should be able to get this merged shortly. |
Description
This adds all four current Llama 3 models and enables Llama-2-70b-hf. No dependencies required for this change (but you must have been granted access on Hugging Face to download).
Type of change
Checklist:
I didn't add tests but I did write a sanity check:
test.py
:output: