Llama 3 support?

[joshmackwilliams]

Is Llama 3 (specifically the instruct models) supported by the code as-is (Llama 2 path), or are changes required (e.g., to support the new tokenizer)?

And, do you have any plans to release a pretrained Eagle for Llama 3? (e.g., like https://huggingface.co/yuhuili/EAGLE-llama2-chat-7B)

[Liyuhui-12]

You need to modify the instruction templates (such as those in eagle/ge_data/ge_data_all_vicuna.py). Training the draft model for LLaMA 3 is our next step.

[kalradivyanshu]

Hey @Liyuhui-12 any updates on LLaMA-3 support. Btw, thank you for all your hard work!

[Liyuhui-12]

Hey @kalradivyanshu, support for LLaMA3 has now been updated.