Summary of Openbezoar: Small, Cost-effective and Open Models Trained on Mixes Of Instruction Data, by Chandeepa Dissanayake et al.
OpenBezoar: Small, Cost-Effective and Open Models Trained on Mixes of Instruction Data
by Chandeepa Dissanayake, Lahiru Lowe, Sachith Gunasekara, Yasiru Ratnayake
First submitted to arxiv on: 18 Apr 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 The authors of this paper present a recipe for fine-tuning large language models (LLMs) to perform well on various downstream tasks. They start by generating synthetic instruction fine-tuning data using three different schemes based on existing models, then filter these generations using GPT-4 as a human proxy. Next, they perform cost-effective supervised fine-tuning sequentially with each scheme and further fine-tune the resulting checkpoint with a subset of the HH-RLHF dataset to minimize distribution shift. The final checkpoint, “OpenBezoar-HH-RLHF-DPO”, is evaluated using the LM Eval Harness tasks/metrics as well as on MT-Bench with Claude 2.1, demonstrating superior performance over many models at the 3B parameter scale. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us make better language models that can perform well on different tasks. The authors use a combination of existing models and techniques to fine-tune their model, “OpenBezoar-HH-RLHF-DPO”, which does very well compared to other models at the same level. They also share their code and data so others can try it out. |
Keywords
» Artificial intelligence » Claude » Fine tuning » Gpt » Rlhf » Supervised
