Summary of Improving Small-scale Large Language Models Function Calling For Reasoning Tasks, by Graziano A. Manduzio et al.
Improving Small-Scale Large Language Models Function Calling for Reasoning Tasks
by Graziano A. Manduzio, Federico A. Galatolo, Mario G. C. A. Cimino, Enzo Pasquale Scilingo, Lorenzo Cominelli
First submitted to arxiv on: 24 Oct 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
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 study proposes a novel framework to improve the performances of small-scale Large Language Models (LLMs) in logical and mathematical reasoning tasks. The approach focuses on training smaller LLMs using function calling abilities, allowing them to execute provided functions and utilize their outputs for task completion. The researchers employ an agent that injects descriptions and examples of usable functions into prompts, managing step-by-step reasoning chains. A dataset is created from a large-scale LLM’s correct and incorrect chat completions, which is then used to train smaller LLMs using Reinforcement Learning from Human Feedback (RLHF) with Direct Preference Optimization (DPO). Experimental results demonstrate the proposed approach balances model size and performance, enhancing function calling capabilities for reasoning tasks in smaller models. The study contributes to addressing limitations of LLMs in mathematical problem-solving and logical reasoning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research helps small language models become better at solving math problems and making logical decisions. It’s a new way to teach these models how to use functions, which are like recipes for solving problems. The model is taught by an agent that gives it examples of how to solve problems step-by-step. This training data comes from a bigger model that tried to solve the same problems, but sometimes got them wrong. By using this training data and a special technique called Reinforcement Learning from Human Feedback, the small model learns to make better decisions. The results show that this approach helps smaller models do better at solving math problems and making logical decisions. |
Keywords
» Artificial intelligence » Optimization » Reinforcement learning from human feedback » Rlhf
