English

Transferable Post-training via Inverse Value Learning

Machine Learning 2025-06-16 v2 Computation and Language

Abstract

As post-training processes utilize increasingly large datasets and base models continue to grow in size, the computational demands and implementation challenges of existing algorithms are escalating significantly. In this paper, we propose modeling the changes at the logits level during post-training using a separate neural network (i.e., the value network). After training this network on a small base model using demonstrations, this network can be seamlessly integrated with other pre-trained models during inference, enables them to achieve similar capability enhancements. We systematically investigate the best practices for this paradigm in terms of pre-training weights and connection schemes. We demonstrate that the resulting value network has broad transferability across pre-trained models of different parameter sizes within the same family, models undergoing continuous pre-training within the same family, and models with different vocabularies across families. In certain cases, it can achieve performance comparable to full-parameter fine-tuning. Furthermore, we explore methods to enhance the transferability of the value model and prevent overfitting to the base model used during training.

Keywords

Cite

@article{arxiv.2410.21027,
  title  = {Transferable Post-training via Inverse Value Learning},
  author = {Xinyu Lu and Xueru Wen and Yaojie Lu and Bowen Yu and Hongyu Lin and Haiyang Yu and Le Sun and Xianpei Han and Yongbin Li},
  journal= {arXiv preprint arXiv:2410.21027},
  year   = {2025}
}

Comments

NAACL 2025 Camera Ready

R2 v1 2026-06-28T19:38:02.105Z