File size: 802 Bytes
c76f737 232bece c76f737 232bece |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 |
---
license: apache-2.0
datasets:
- datajuicer/alpaca-cot-en-refined-by-data-juicer
---
This is a reference LLM from [Data-Juicer](https://github.com/alibaba/data-juicer).
The model architecture is LLaMA-7B and we built it upon the pre-trained [checkpoint](https://huggingface.co/huggyllama/llama-7b).
The model is fine-trained on 40k English chat samples of Data-Juicer's refined [alpaca-CoT data](https://github.com/alibaba/data-juicer/blob/main/configs/data_juicer_recipes/alpaca_cot/README.md#refined-alpaca-cot-dataset-meta-info).
It beats LLaMA-7B fine-tuned on 52k Alpaca samples in GPT-4 evaluation.
For more details, please refer to our [paper](https://arxiv.org/abs/2309.02033).
 |