Summary of 52b to 1t: Lessons Learned Via Tele-flm Series, by Xiang Li et al.
52B to 1T: Lessons Learned via Tele-FLM Series
by Xiang Li, Yiqun Yao, Xin Jiang, Xuezhi Fang, Chao Wang, Xinzhang Liu, Zihan Wang, Yu Zhao, Xin Wang, Yuyao Huang, Shuangyong Song, Yongxiang Li, Zheng Zhang, Bo Zhao, Aixin Sun, Yequan Wang, Zhongjiang He, Zhongyuan Wang, Xuelong Li, Tiejun Huang
First submitted to arxiv on: 3 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 This technical report explores the potential of Large Language Models (LLMs) in Artificial General Intelligence. Building on our previous work with Tele-FLM-52B, a 52-billion-parameter model, we investigate two key areas: Supervised Fine-tuning (SFT) on Tele-FLM-52B and best practices for scaling up the model to 102 billion parameters and eventually 1 trillion. Our findings support the “less is more” approach for SFT data construction and demonstrate experimentally validated strategies for progressively growing a large language model. We will open-source our 1T model checkpoint, Tele-FLM-1T, to facilitate further research and training. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper takes a big step towards creating super smart computers that can understand any language. We’re building on previous work with a special type of computer model called Tele-FLM-52B. We want to know how well this model works when we make it even bigger and better. Our research shows that sometimes, less is more, especially when teaching the model new things. We also figured out how to make the model grow gradually from 52 billion parameters to 1 trillion. To help others learn from our findings, we’ll release a special copy of our biggest and best model yet, called Tele-FLM-1T. |
Keywords
» Artificial intelligence » Fine tuning » Large language model » Supervised
