Automatic Pair Construction for Contrastive Post-training

TL;DR

This paper introduces an automatic method for constructing contrastive data from multiple LLMs to improve alignment, demonstrating significant performance gains and surpassing ChatGPT in instruction tuning.

Contribution

It proposes an automatic contrastive data construction method using multiple models and a curriculum learning scheme, enhancing LLM alignment beyond existing supervised fine-tuning.

Findings

DPO outperforms SFT baselines significantly.

Curriculum learning improves alignment results.

Scaling to larger models like Orca yields state-of-the-art performance.

Abstract

Alignment serves as an important step to steer large language models (LLMs) towards human preferences. In this paper, we propose an automatic way to construct contrastive data for LLM, using preference pairs from multiple models of varying strengths (e.g., InstructGPT, ChatGPT and GPT-4). We compare the contrastive techniques of SLiC and DPO to SFT baselines and find that DPO provides a step-function improvement even after continuing SFT saturates. We also explore a data curriculum learning scheme for contrastive post-training, which starts by learning from "easier" pairs and transitioning to "harder" ones, which further improves alignment. Finally, we scale up our experiments to train with more data and larger models like Orca. Remarkably, our automatic contrastive post-training further improves the performance of Orca, already a state-of-the-art instruction learning model tuned with GPT-4 outputs, to outperform ChatGPT.