Saved in:
Bibliographic Details
Main Authors: Ni, Bolin, Hu, JingCheng, Wei, Yixuan, Peng, Houwen, Zhang, Zheng, Meng, Gaofeng, Hu, Han
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2405.20335
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910464560070656
author Ni, Bolin
Hu, JingCheng
Wei, Yixuan
Peng, Houwen
Zhang, Zheng
Meng, Gaofeng
Hu, Han
author_facet Ni, Bolin
Hu, JingCheng
Wei, Yixuan
Peng, Houwen
Zhang, Zheng
Meng, Gaofeng
Hu, Han
contents In this work, we present Xwin-LM, a comprehensive suite of alignment methodologies for large language models (LLMs). This suite encompasses several key techniques, including supervised finetuning (SFT), reward modeling (RM), rejection sampling finetuning (RS), and direct preference optimization (DPO). The key components are as follows: (1) Xwin-LM-SFT, models initially finetuned with high-quality instruction data; (2) Xwin-Pair, a large-scale, multi-turn preference dataset meticulously annotated using GPT-4; (3) Xwin-RM, reward models trained on Xwin-Pair, developed at scales of 7B, 13B, and 70B parameters; (4) Xwin-Set, a multiwise preference dataset in which each prompt is linked to 64 unique responses generated by Xwin-LM-SFT and scored by Xwin-RM; (5) Xwin-LM-RS, models finetuned with the highest-scoring responses from Xwin-Set; (6) Xwin-LM-DPO, models further optimized on Xwin-Set using the DPO algorithm. Our evaluations on AlpacaEval and MT-bench demonstrate consistent and significant improvements across the pipeline, demonstrating the strength and scalability of Xwin-LM. The repository https://github.com/Xwin-LM/Xwin-LM will be continually updated to foster community research.
format Preprint
id arxiv_https___arxiv_org_abs_2405_20335
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Xwin-LM: Strong and Scalable Alignment Practice for LLMs
Ni, Bolin
Hu, JingCheng
Wei, Yixuan
Peng, Houwen
Zhang, Zheng
Meng, Gaofeng
Hu, Han
Computation and Language
In this work, we present Xwin-LM, a comprehensive suite of alignment methodologies for large language models (LLMs). This suite encompasses several key techniques, including supervised finetuning (SFT), reward modeling (RM), rejection sampling finetuning (RS), and direct preference optimization (DPO). The key components are as follows: (1) Xwin-LM-SFT, models initially finetuned with high-quality instruction data; (2) Xwin-Pair, a large-scale, multi-turn preference dataset meticulously annotated using GPT-4; (3) Xwin-RM, reward models trained on Xwin-Pair, developed at scales of 7B, 13B, and 70B parameters; (4) Xwin-Set, a multiwise preference dataset in which each prompt is linked to 64 unique responses generated by Xwin-LM-SFT and scored by Xwin-RM; (5) Xwin-LM-RS, models finetuned with the highest-scoring responses from Xwin-Set; (6) Xwin-LM-DPO, models further optimized on Xwin-Set using the DPO algorithm. Our evaluations on AlpacaEval and MT-bench demonstrate consistent and significant improvements across the pipeline, demonstrating the strength and scalability of Xwin-LM. The repository https://github.com/Xwin-LM/Xwin-LM will be continually updated to foster community research.
title Xwin-LM: Strong and Scalable Alignment Practice for LLMs
topic Computation and Language
url https://arxiv.org/abs/2405.20335