File size: 1,765 Bytes
79e7dcb 1eb1bba 79e7dcb 1eb1bba 79e7dcb 1eb1bba 79e7dcb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 |
---
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}
}
``` |