|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- databricks/databricks-dolly-15k |
|
language: |
|
- en |
|
metrics: |
|
- rouge |
|
base_model: |
|
- openai-community/gpt2-large |
|
pipeline_tag: text-generation |
|
--- |
|
|
|
|
|
# MiniLLM-gpt2-760M |
|
|
|
[paper](https://arxiv.org/abs/2306.08543) | [code](https://github.com/microsoft/LMOps/tree/main/minillm) |
|
|
|
**MiniLLM-gpt2-760M** is a gpt2-large (760M) model distilled from [gpt2-xlarge (1.5B)](https://huggingface.co/MiniLLM/teacher-gpt2-1.5B) on [databricks-dolly-15k](https://huggingface.co/datasets/aisquared/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](https://huggingface.co/MiniLLM/init-gpt2-760M) 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](https://huggingface.co/datasets/aisquared/databricks-dolly-15k) (test set), [self-instruct](https://github.com/tatsu-lab/stanford_alpaca/blob/main/alpaca_data.json), and [vicuna](https://github.com/lm-sys/vicuna-blog-eval) |
|
|
|
<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](https://huggingface.co/MiniLLM/SFT-gpt2-760M) |
|
+ [KD](https://huggingface.co/MiniLLM/KD-gpt2-760M) |
|
+ [SeqKD](https://huggingface.co/MiniLLM/SeqKD-gpt2-760M) |
|
|
|
## 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} |
|
} |
|
``` |
