Tuna: Instruction Tuning using Feedback from Large Language Models

TL;DR

Tuna is a novel instruction tuning method that leverages feedback from large language models using probabilistic and contextual ranking to improve response quality across multiple benchmarks.

Contribution

The paper introduces Tuna, a new approach for instruction tuning that uses ranking techniques to incorporate feedback from powerful LLMs, enhancing response quality.

Findings

Tuna outperforms several reinforcement learning baselines.

It improves performance on multiple benchmarks.

The method effectively incorporates LLM feedback.

Abstract

Instruction tuning of open-source large language models (LLMs) like LLaMA, using direct outputs from more powerful LLMs such as Instruct-GPT and GPT-4, has proven to be a cost-effective way to align model behaviors with human preferences. However, the instruction-tuned model has only seen one response per instruction, lacking the knowledge of potentially better responses. In this paper, we propose finetuning an instruction-tuned LLM using our novel \textit{probabilistic ranking} and \textit{contextual ranking} approaches to increase the likelihood of generating better responses. Probabilistic ranking enables the instruction-tuned model to inherit the relative rankings of high-quality and low-quality responses from the teacher LLM. On the other hand, learning with contextual ranking allows the model to refine its own response distribution using the contextual understanding ability of stronger LLMs. Furthermore, we apply probabilistic ranking and contextual ranking sequentially to the instruction-tuned LLM. The resulting model, which we call \textbf{Tuna}, consistently improves the performance on Super Natural Instructions (119 test tasks), LMentry (25 test tasks), Vicuna QA, and can even obtain better results than several strong reinforcement learning baselines. Our code and data are available at \url{ https://github.com/microsoft/LMOps}.