Open cduk opened 1 month ago
Might be easier to train a medusa head?
On Wed, May 22, 2024 at 11:01 PM cduk @.***> wrote:
Given that we have only Llama 3 70B and 8B, it would be useful to have a Tiny Llama based on the Llama 3 tokenizer so that we can use it as a drafting model for speculative decoding.
Are there plans to create a Llama 3 version?
— Reply to this email directly, view it on GitHub https://github.com/jzhang38/TinyLlama/issues/186, or unsubscribe https://github.com/notifications/unsubscribe-auth/ASVG6CXCKDERGD5SJZALQQLZDUISXAVCNFSM6AAAAABIEO26WKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGMYTCNJWHE3TONY . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Given that we have only Llama 3 70B and 8B, it would be useful to have a Tiny Llama based on the Llama 3 tokenizer so that we can use it as a drafting model for speculative decoding.
Are there plans to create a Llama 3 version?