Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Zhichao, Bi, Bin, Zhu, Zixu, Mao, Xiangbo, Wang, Jun, Wang, Shiyu, Wang, Cheng, Nie, Dong, Hong, Lingzi
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2410.21438
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913099741659136
author Wang, Zhichao
Bi, Bin
Zhu, Zixu
Mao, Xiangbo
Wang, Jun
Wang, Shiyu
Wang, Cheng
Nie, Dong
Hong, Lingzi
author_facet Wang, Zhichao
Bi, Bin
Zhu, Zixu
Mao, Xiangbo
Wang, Jun
Wang, Shiyu
Wang, Cheng
Nie, Dong
Hong, Lingzi
contents By pretraining on trillions of tokens, an LLM gains the capability of text generation. However, to enhance its utility and reduce potential harm, SFT and alignment are applied sequentially to the pretrained model. Because SFT and alignment have different objectives and underlying processes, performance on certain tasks can decline. To address this, we seamlessly introduce Unified Fine-Tuning (UFT), which integrates SFT and alignment into a single training stage using the same objective and loss functions through an implicit reward function. Our experimental results demonstrate that UFT outperforms SFT on instruction-tuning data alone. Moreover, when combining instruction-tuning data with alignment data, UFT effectively prevents the degradation on some tasks across these two stages and shows a clear advantage over sequentially applying SFT and alignment. This is evident in the significant improvements observed in the \textbf{ifeval} task for instruction-following and the \textbf{truthful} task for factuality. The proposed general fine-tuning framework UFT establishes an effective and efficient paradigm for LLM post-training.
format Preprint
id arxiv_https___arxiv_org_abs_2410_21438
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle UFT: Unifying Fine-Tuning of SFT and RLHF/DPO/UNA through a Generalized Implicit Reward Function
Wang, Zhichao
Bi, Bin
Zhu, Zixu
Mao, Xiangbo
Wang, Jun
Wang, Shiyu
Wang, Cheng
Nie, Dong
Hong, Lingzi
Computation and Language
Machine Learning
By pretraining on trillions of tokens, an LLM gains the capability of text generation. However, to enhance its utility and reduce potential harm, SFT and alignment are applied sequentially to the pretrained model. Because SFT and alignment have different objectives and underlying processes, performance on certain tasks can decline. To address this, we seamlessly introduce Unified Fine-Tuning (UFT), which integrates SFT and alignment into a single training stage using the same objective and loss functions through an implicit reward function. Our experimental results demonstrate that UFT outperforms SFT on instruction-tuning data alone. Moreover, when combining instruction-tuning data with alignment data, UFT effectively prevents the degradation on some tasks across these two stages and shows a clear advantage over sequentially applying SFT and alignment. This is evident in the significant improvements observed in the \textbf{ifeval} task for instruction-following and the \textbf{truthful} task for factuality. The proposed general fine-tuning framework UFT establishes an effective and efficient paradigm for LLM post-training.
title UFT: Unifying Fine-Tuning of SFT and RLHF/DPO/UNA through a Generalized Implicit Reward Function
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2410.21438