MiniLLM/MiniLLM-gpt2-760M

MiniLLM-gpt2-760M

paper | code

MiniLLM-gpt2-760M is a gpt2-large (760M) model distilled from gpt2-xlarge (1.5B) on databricks-dolly-15k

<p align='left'> <img src="https://cdn-uploads.huggingface.co/production/uploads/624ac662102fcdff87be51b9/7hBWGZzYMJihCRQ70XoiQ.png" width="1000"> </p>

Note: MiniLLM requires a SFT model for initilization to perform the PPO optimization.

Evaluation

We ask GPT-4 to give scores for the generated responses of MiniLLM. The prompts are taken from databricks-dolly-15k (test set), self-instruct, and vicuna

<p align='left'> <img src="https://cdn-uploads.huggingface.co/production/uploads/624ac662102fcdff87be51b9/rDXnaDbKH5mBYAmqGC-_a.png" width="1000"> </p>

Baseline Models

• —SFT w/o KD
• —KD
• —SeqKD

Citation

@inproceedings{minillm,
  title={MiniLLM: Knowledge Distillation of Large Language Models},
  author={Gu, Yuxian and Dong, Li and Wei, Furu and Huang, Minlie},
  booktitle={Proceedings of ICLR},
  year={2024}
}