AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback

TL;DR

AlpacaFarm is a low-cost simulation framework for developing and evaluating methods that learn from human feedback in training large language models, enabling efficient research and benchmarking.

Contribution

The paper introduces AlpacaFarm, a simulation environment that reduces data collection costs, provides trustworthy automatic evaluation, and offers reference implementations for feedback-based learning methods.

Findings

Reward model methods outperform supervised fine-tuning.

PPO implementation achieves +10% win-rate against Davinci003.

Model rankings in AlpacaFarm align with real human feedback.

Abstract

Large language models (LLMs) such as ChatGPT have seen widespread adoption due to their strong instruction-following abilities. Developing these LLMs involves a complex yet poorly understood workflow requiring training with human feedback. Replicating and understanding this instruction-following requires tackling three major challenges: the high cost of data collection, the lack of trustworthy evaluation, and the absence of reference method implementations. We address these challenges with AlpacaFarm, a simulator that enables research and development for learning from feedback at a low cost. First, we design LLM prompts to simulate human feedback that are 50x cheaper than crowdworkers and display high agreement with humans. Second, we propose an automatic evaluation and validate it against human instructions obtained on real-world interactions. Third, we contribute reference implementations for several methods (PPO, DPO, best-of-n, expert iteration, and more) that learn from pairwise feedback. Finally, as an end-to-end validation of AlpacaFarm, we train and evaluate eleven models on 10k pairs of real human feedback and show that rankings of models trained in AlpacaFarm match rankings of models trained on human data. As a demonstration of the research possible in AlpacaFarm, we find that methods that use a reward model can substantially improve over supervised fine-tuning and that our reference PPO implementation leads to a +10% improvement in win-rate against Davinci003. We release all components of AlpacaFarm at https://github.com/tatsu-lab/alpaca_farm.