Escaping Collapse: The Strength of Weak Data for Large Language Model Training
This addresses the challenge of ensuring stable LLM training with synthetic data, which is incremental as it builds on boosting techniques and existing methods.
The paper tackles the problem of synthetic data causing performance collapse in large language models by developing a theoretical framework showing minimal curation is needed for continual improvement, and demonstrates a training procedure that converges to optimal performance even with mostly poor-quality non-synthetic data.
Synthetically-generated data plays an increasingly larger role in training large language models. However, while synthetic data has been found to be useful, studies have also shown that without proper curation it can cause LLM performance to plateau, or even "collapse", after many training iterations. In this paper, we formalize this question and develop a theoretical framework to investigate how much curation is needed in order to ensure that LLM performance continually improves. We find that the requirements are nearly minimal. We describe a training procedure that converges to an optimal LLM even if almost all of the non-synthetic training data is of poor quality. Our analysis is inspired by boosting, a classic machine learning technique that leverages a very weak learning algorithm to produce an arbitrarily good classifier. Our training procedure subsumes many recently proposed methods for training LLMs on synthetic data, and thus our analysis sheds light on why they are successful, and also suggests opportunities for future improvement. We present experiments that validate our theory, and show that dynamically focusing labeling resources on the most challenging examples -- in much the same way that boosting focuses the efforts of the weak learner -- leads to improved performance.