wantsleep/granite-3.1-1b-a400m-EAGLE3

granite 모델은 attention multiplier, embedding multiplier등 특별한 변수들이 있어서, eagle inference에도 그 변수들을 적용해주어야한다. 그래서 무조건 SafeAILab/EAGLE github이 아니라, "https://github.com/seohyunwoo-0407/EAGLE_granite" 레포지토리에서 코드를 실행해야한다.

Granite models have special variables such as an attention multiplier and an embedding multiplier, so those must also be applied during EAGLE inference. Therefore, we must run the code from the https://github.com/seohyunwoo-0407/EAGLE_granite repository, not from the official SafeAILab/EAGLE GitHub repository.