Summary of Phudge: Phi-3 As Scalable Judge, by Mahesh Deshwal et al.
PHUDGE: Phi-3 as Scalable Judge
by Mahesh Deshwal, Apoorva Chawla
First submitted to arxiv on: 12 May 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 presents PHUDGE, a fine-tuned Phi3 model that achieves state-of-the-art (SOTA) results in four tasks: Feedback Test, Feedback OOD, MT Human, and Preference Test. PHUDGE surpasses existing models in latency and throughput, demonstrating strong correlation with both GPT4 and human annotators on unseen data. The authors address the use of small language models for cost-effective production-grade systems and show that causal modeling can hinder model learning capabilities. By employing systematic ML experimentation, thoughtful data augmentation, and problem re-purposing, PHUDGE even beats larger models with less training data. The paper also introduces a generalized version of Earth Movers Distance (Wasserstein distance) using Minkowski Distance with a penalty to control loss smoothing. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research shows that a special type of AI model called PHUDGE can perform four different tasks better than any other existing models. These tasks include giving feedback, understanding language, and making decisions based on human preferences. PHUDGE is able to do this quickly and efficiently, using less training data than bigger models. The authors also introduce a new way to measure how well an AI model is doing, which helps them train the model better. |
Keywords
» Artificial intelligence » Data augmentation
