Context-Free Synthetic Data Mitigates Forgetting
This addresses forgetting in fine-tuning for language models, offering a practical solution for settings with limited data access, though it is incremental as it builds on existing KL divergence methods.
The paper tackles the problem of catastrophic forgetting in language models during fine-tuning by proposing context-free generation to estimate KL divergence, showing it mitigates forgetting and preserves zero-shot and reasoning performance better than contextual synthetic data or pretraining data, with results on models like OLMo-1B and R1-Distill-Llama-8B.
Fine-tuning a language model often results in a degradation of its existing performance on other tasks, due to a shift in the model parameters; this phenomenon is often referred to as (catastrophic) forgetting. We are interested in mitigating this, in settings where we only have access to the model weights but no access to its training data/recipe. A natural approach is to penalize the KL divergence between the original model and the new one. Our main realization is that a simple process - which we term context-free generation - allows for an approximate unbiased estimation of this KL divergence. We show that augmenting a fine-tuning dataset with context-free generations mitigates forgetting, in two settings: (a) preserving the zero-shot performance of pretrained-only models, and (b) preserving the reasoning performance of thinking models. We show that contextual synthetic data, and even a portion of the pretraining data, are less effective. We also investigate the effect of choices like generation temperature, data ratios etc. We present our results for OLMo-1B for pretrained-only setting and R1-Distill-Llama-8B for the reasoning setting.