LaFFi: Leveraging Hybrid Natural Language Feedback for Fine-tuning Language Models
This addresses performance issues in fine-tuning LLMs for specific tasks, but it is incremental as it builds on existing SFT methods.
The paper tackles the problem of hallucinations and mistakes in LLMs fine-tuned with SFT by introducing LaFFi, which uses natural language feedback to improve accuracy in question-answering tasks, showing significant gains.
Fine-tuning Large Language Models (LLMs) adapts a trained model to specific downstream tasks, significantly improving task-specific performance. Supervised Fine-Tuning (SFT) is a common approach, where an LLM is trained to produce desired answers. However, LLMs trained with SFT sometimes make simple mistakes and result in hallucinations on reasoning tasks such as question-answering. Without external feedback, it is difficult for SFT to learn a good mapping between the question and the desired answer, especially with a small dataset. This paper introduces an alternative to SFT called Natural Language Feedback for Finetuning LLMs (LaFFi). LaFFi has LLMs directly predict the feedback they will receive from an annotator. We find that requiring such reflection can significantly improve the accuracy in in-domain question-answering tasks, providing a promising direction for the application of natural language feedback in the realm of SFT LLMs. Additional ablation studies show that the portion of human-annotated data in the annotated datasets affects the fine-tuning performance.