https://arxiv.org/abs/2308.08998

Skip to main content
Cornell University
We gratefully acknowledge support from the Simons Foundation, member
institutions, and all contributors. Donate
 
arxiv logo > cs > arXiv:2308.08998
[                    ]

Help | Advanced Search

[All fields        ]
Search
arXiv logo
Cornell University Logo
[                    ] GO
quick links

  * Login
  * Help Pages
  * About

Computer Science > Computation and Language

arXiv:2308.08998 (cs)
[Submitted on 17 Aug 2023 (v1), last revised 21 Aug 2023 (this
version, v2)]

Title:Reinforced Self-Training (ReST) for Language Modeling

Authors:Caglar Gulcehre, Tom Le Paine, Srivatsan Srinivasan, Ksenia
Konyushkova, Lotte Weerts, Abhishek Sharma, Aditya Siddhant, Alex
Ahern, Miaosen Wang, Chenjie Gu, Wolfgang Macherey, Arnaud Doucet,
Orhan Firat, Nando de Freitas
Download a PDF of the paper titled Reinforced Self-Training (ReST)
for Language Modeling, by Caglar Gulcehre and 13 other authors
Download PDF

    Abstract: Reinforcement learning from human feedback (RLHF) can
    improve the quality of large language model's (LLM) outputs by
    aligning them with human preferences. We propose a simple
    algorithm for aligning LLMs with human preferences inspired by
    growing batch reinforcement learning (RL), which we call
    Reinforced Self-Training (ReST). Given an initial LLM policy,
    ReST produces a dataset by generating samples from the policy,
    which are then used to improve the LLM policy using offline RL
    algorithms. ReST is more efficient than typical online RLHF
    methods because the training dataset is produced offline, which
    allows data reuse. While ReST is a general approach applicable to
    all generative learning settings, we focus on its application to
    machine translation. Our results show that ReST can substantially
    improve translation quality, as measured by automated metrics and
    human evaluation on machine translation benchmarks in a compute
    and sample-efficient manner.

Comments: 23 pages, 16 figures
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as:  arXiv:2308.08998 [cs.CL]
          (or arXiv:2308.08998v2 [cs.CL] for this version)
          https://doi.org/10.48550/arXiv.2308.08998
          Focus to learn more
          arXiv-issued DOI via DataCite

Submission history

From: Caglar Gulcehre [view email]
[v1] Thu, 17 Aug 2023 14:12:48 UTC (2,542 KB)
[v2] Mon, 21 Aug 2023 10:23:42 UTC (1,040 KB)
Full-text links:

Download:

  * Download a PDF of the paper titled Reinforced Self-Training
    (ReST) for Language Modeling, by Caglar Gulcehre and 13 other
    authors
    PDF
  * Other formats

[by-nc-nd-4]
Current browse context:
cs.CL
< prev   |   next >
new | recent | 2308
Change to browse by:
cs
cs.LG

References & Citations

  * NASA ADS
  * Google Scholar
  * Semantic Scholar

a export BibTeX citation Loading...

BibTeX formatted citation

x
[loading...          ]
Data provided by:

Bookmark

BibSonomy logo Reddit logo
(*) Bibliographic Tools

Bibliographic and Citation Tools

[ ] Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
[ ] Litmaps Toggle
Litmaps (What is Litmaps?)
[ ] scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
( ) Code, Data, Media

Code, Data and Media Associated with this Article

[ ] Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
[ ] DagsHub Toggle
DagsHub (What is DagsHub?)
[ ] Links to Code Toggle
Papers with Code (What is Papers with Code?)
[ ] ScienceCast Toggle
ScienceCast (What is ScienceCast?)
( ) Demos

Demos

[ ] Replicate Toggle
Replicate (What is Replicate?)
[ ] Spaces Toggle
Hugging Face Spaces (What is Spaces?)
( ) Related Papers

Recommenders and Search Tools

[ ] Link to Influence Flower
Influence Flower (What are Influence Flowers?)
[ ] Connected Papers Toggle
Connected Papers (What is Connected Papers?)
[ ] Core recommender toggle
CORE Recommender (What is CORE?)

  * Author
  * Venue
  * Institution
  * Topic

( ) About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and
share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have
embraced and accepted our values of openness, community, excellence,
and user data privacy. arXiv is committed to these values and only
works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community?
Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is
MathJax?)

  * About
  * Help

  * Click here to contact arXiv Contact
  * Click here to subscribe Subscribe

  * Copyright
  * Privacy Policy

  * Web Accessibility Assistance
  * arXiv Operational Status
    Get status notifications via email or slack