Summary of Fox-1 Technical Report, by Zijian Hu et al.
Fox-1 Technical Report
by Zijian Hu, Jipeng Zhang, Rui Pan, Zhaozhuo Xu, Shanshan Han, Han Jin, Alay Dilipbhai Shah, Dimitris Stripelis, Yuhang Yao, Salman Avestimehr, Chaoyang He, Tong Zhang
First submitted to arxiv on: 8 Nov 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The Fox-1 series of small language models (SLMs) is introduced as a novel approach in natural language processing. The pre-trained models, consisting of Fox-1-1.6B and its variant Fox-1-1.6B-Instruct-v0.1, utilize 3 trillion tokens of web-scraped document data and fine-tune with 5 billion tokens of instruction-following and multi-turn conversation data. The innovative design features a novel 3-stage data curriculum, Grouped Query Attention (GQA), and an expanded vocabulary. This architecture leads to improved performance in various benchmarks compared to existing SLMs like StableLM-2-1.6B, Gemma-2B, Qwen1.5-1.8B, and OpenELM1.1B. The model’s weights are released under the Apache 2.0 license, aiming to democratize large language models (LLMs) and make them accessible to the open-source community. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Fox-1 is a new way to build small language models that can understand and generate human-like text. It uses lots of data from the internet and teaches itself to follow instructions and have conversations. The model has three stages of training, which helps it learn faster and better than other similar models. Fox-1 also has a special attention mechanism called Grouped Query Attention, which makes it more efficient and effective. The model is tested on several benchmarks and performs as well or even better than some existing models. The authors want to share their model with the world by releasing its weights under an open-source license. |
Keywords
» Artificial intelligence » Attention » Natural language processing
