meta-llama

/

Llama-3.2-1B-Instruct

like 540

Follow

Meta Llama 7,920

My result:
micro average is 46.31
macro average across categories is 46.67
macro average across sub-categories is 48.2

I used this code "https://github.com/meta-llama/llama3"

Could it be because FlashAttention is not used in this code, resulting in a difference from the evaluation results on Hugging Face?