Question about Llama 3.1 license compliance

#3
by tokinasin - opened

Hello,

I have a quick question about the license for this model.

The model tree states this is a fine-tune of meta-llama/Llama-3.1-8B. However, this repository does not provide a copy of Llama 3.1 license.

Is there a specific agreement between Microsoft and Meta that allows for this situation?

If not, Llama 3.1 license policy seems to require that derivative models:

  1. Include a copy of the Meta Llama 3.1 Community License.
  2. Include "Llama" in the model name.
  3. Include the "Built with Llama" attribution.

Could you please clarify the situation?

Thanks for your great work.

Well, the paper mentions they fine-tuned "Llama3-8b-Base". Not sure what model this is actually... could be Llama 3 (which doesn't answer the license questions and makes the situation even more unclear).

On the model card under "Training procedure ", they say:

We performed full-parameter fine-tuning of Llama3-8b-Base

The guff information says:

{F1CF7897-98BC-4B0B-ABE3-041476784427}

In either case, the license is Llama3's license and not MIT as listed on the huggingface page.
They aren't adhering to Llama's license.

Sign up or log in to comment