Summary of From Language Models Over Tokens to Language Models Over Characters, by Tim Vieira et al.
From Language Models over Tokens to Language Models over Characters
by Tim Vieira, Ben LeBrun, Mario Giulianelli, Juan Luis Gastaldi, Brian DuSell, John Terilla, Timothy J. O’Donnell, Ryan Cotterell
First submitted to arxiv on: 4 Dec 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 paper addresses a significant challenge in building user applications on top of modern language models, which are internally distributions over token strings rather than character strings. The problem arises when a prompt needs to be tokenized before passing it to the token-level language model, making the tokenizer and subsequent analyses sensitive to the prompt specification. To overcome this limitation, the authors present algorithms for converting token-level language models to character-level ones, including exact and approximate methods. The empirical evaluation shows that even with a limited computation budget, the proposed method can accurately approximate the character-level distribution (less than 0.00021 excess bits per character) at reasonable speeds (46.3 characters per second) on the Llama 3.1 8B language model. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you have a super smart computer program that can understand and generate human-like text, but it only works with special “tokens” rather than regular letters and spaces. This makes it hard to use the program in certain situations because the way you ask the question or phrase it matters a lot. The authors of this paper found ways to fix this problem by creating new algorithms that can turn the token-based language model into one that works with regular characters. They tested their methods on a specific type of language model and showed that they work really well, even when the computer has limited power. |
Keywords
» Artificial intelligence » Language model » Llama » Prompt » Token » Tokenizer
