Summary of Chared: Character-wise Ensemble Decoding For Large Language Models, by Kevin Gu et al.
CharED: Character-wise Ensemble Decoding for Large Language Models
by Kevin Gu, Eva Tuecke, Dmitriy Katz, Raya Horesh, David Alvarez-Melis, Mikhail Yurochkin
First submitted to arxiv on: 25 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 This paper presents a novel inference-time ensembling algorithm for combining the outputs of large language models (LLMs) without requiring shared vocabularies or tokenizations. The proposed method, called Character-wise Ensemble Decoding (CharED), averages the marginal distributions of each character across multiple LLMs to generate an output, character by character. Experimental results show that CharED improves performance across multiple domains, including coding, math, and toxicity benchmarks, by combining the strengths of individual models regardless of their sizes or tokenizations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about a new way to use different language models together. Right now, we have many big language models that are really good at solving problems. But when we want to use multiple models together, it can be tricky because they might not understand the same words or symbols. The researchers found a solution by creating an algorithm called CharED. It works by taking the output from each model and averaging them out, one character at a time, to get a better answer. They tested this method on different tasks like coding and math problems, and it worked really well. |
Keywords
* Artificial intelligence * Inference
