Summary of Self-refinement Of Language Models From External Proxy Metrics Feedback, by Keshav Ramji et al.
Self-Refinement of Language Models from External Proxy Metrics Feedback
by Keshav Ramji, Young-Suk Lee, Ramón Fernandez Astudillo, Md Arafat Sultan, Tahira Naseem, Asim Munawar, Radu Florian, Salim Roukos
First submitted to arxiv on: 27 Feb 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 This paper introduces Proxy Metric-based Self-Refinement (ProMiSe), a method for Large Language Models (LLMs) to refine their initial responses along key dimensions of quality, guided by external metrics feedback. ProMiSe enables an LLM to iteratively refine its response one principle at a time, leveraging feedback on response quality through principle-specific proxy metrics. The authors apply ProMiSe to open-source language models Flan-T5-XXL and Llama-2-13B-Chat, evaluating performance on document-grounded question answering datasets MultiDoc2Dial and QuAC. Results show that self-refinement improves response quality, and fine-tuning the model on synthetic dialogue data generated by ProMiSe yields significant performance improvements over a zero-shot baseline. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps computers give better answers to questions. It’s like an editor for computer responses! The method is called ProMiSe and it makes sure the answer is good quality by looking at specific things like whether it’s relevant to what was asked and if it matches what’s in a document. The researchers tested this on two big language models and found that it works really well. This means computers can give even better answers to questions, which is useful for things like chatbots and search engines. |
Keywords
* Artificial intelligence * Fine tuning * Llama * Question answering * T5 * Zero shot
