| --- |
| license: apache-2.0 |
| datasets: |
| - databricks/databricks-dolly-15k |
| language: |
| - en |
| metrics: |
| - rouge |
| base_model: |
| - openai-community/gpt2-medium |
| pipeline_tag: text-generation |
| --- |
| # KD-gpt2-340M |
|
|
| [paper](https://arxiv.org/abs/2306.08543) | [code](https://github.com/microsoft/LMOps/tree/main/minillm) |
|
|
| **KD-gpt2-340M** is a gpt2-medium (120M) 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) with token-level forward KLD. |
|
|
| It is used as a baseline for [MiniLLM](https://huggingface.co/MiniLLM/MiniLLM-gpt2-340M). |
|
|
|
|
| ## Other Baselines |
| + [SFT w/o KD](https://huggingface.co/MiniLLM/SFT-gpt2-340M) |
| + [SeqKD](https://huggingface.co/MiniLLM/SeqKD-gpt2-340M) |
|
|
|
|
| ## 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} |
| } |
| ``` |