---
license: apache-2.0
datasets:
- gair-prox/c4-pro
language:
- en
tags:
- llama
---

# C4-ProX-1.7B

<p align="center">
  <img src="prox-teaser.png">
</p>

[ArXiv](http://arxiv.org/abs/2409.17115) | [Models](https://huggingface.co/gair-prox/C4-ProX-1.7B) | [Data](https://huggingface.co/datasets/gair-prox/c4-pro) | [Code](https://github.com/GAIR-NLP/program-every-example)

**C4-ProX-1.7B** is a small language model. It was and trained on the [C4-pro](https://huggingface.co/datasets/gair-prox/c4-pro) for 50B tokens. 

## Evaluations

ProX models are evaluated over 10 language model benchmarks in zero-shot setting.

|                       | ArC-c | ARC-e | CSQA  | HellaS | MMLU  | OBQA  | PiQA  | SIQA  | WinoG | SciQ  | AVG  |
|-----------------------|-------|-------|-------|-----------|-------|-------|-------|-------|-------|-------|------|
| raw | 25.3 | 48.8 | 30.1 | 52.4 | 28.8 | 32.2 | 72.0 | 40.6 | 53.6 | 71.7 | 45.5 |
| ours | 31.1 | 56.0 | 28.4 | 55.2 | 31.1 | 36.2 | 74.0 | 41.0 | 54.1 | 76.8 | 48.4 |

### Citation
```
@article{zhou2024programming,
  title={Programming Every Example: Lifting Pre-training Data Quality like Experts at Scale},
  author={Zhou, Fan and Wang, Zengzhi and Liu, Qian and Li, Junlong and Liu, Pengfei},
  journal={arXiv preprint arXiv:2409.17115},
  year={2024}
}
```