Upweighting Easy Samples in Fine-Tuning Mitigates Forgetting

TL;DR

This paper introduces a sample weighting scheme based on pre-trained model losses to mitigate catastrophic forgetting during fine-tuning, effectively preserving original capabilities while adapting to new tasks.

Contribution

It proposes a novel sample weighting method that upweights easy samples based on pre-trained model losses, addressing forgetting without access to original training data.

Findings

Reduces accuracy drop on target task during fine-tuning.

Preserves more pre-training knowledge compared to standard fine-tuning.

Effective on both language and vision tasks.

Abstract

Fine-tuning a pre-trained model on a downstream task often degrades its original capabilities, a phenomenon known as "catastrophic forgetting". This is especially an issue when one does not have access to the data and recipe used to develop the pre-trained model. Under this constraint, most existing methods for mitigating forgetting are inapplicable. To address this challenge, we propose a sample weighting scheme for the fine-tuning data solely based on the pre-trained model's losses. Specifically, we upweight the easy samples on which the pre-trained model's loss is low and vice versa to limit the drift from the pre-trained model. Our approach is orthogonal and yet complementary to existing methods; while such methods mostly operate on parameter or gradient space, we concentrate on the sample space. We theoretically analyze the impact of fine-tuning with our method in a linear setting, showing that it stalls learning in a certain subspace which inhibits overfitting to the target task. We empirically demonstrate the efficacy of our method on both language and vision tasks. As an example, when fine-tuning Gemma 2 2B on MetaMathQA, our method results in only a $0.8\%$ drop in accuracy on GSM8K (another math dataset) compared to standard fine-tuning, while preserving $5.4\%$ more accuracy on the pre-training datasets. Our code is publicly available at https://github.com/sanyalsunny111/FLOW_finetuning .