https://arxiv.org/abs/2408.16737

close this message
arXiv Accessibility Forum 2024

Skip to main content
Cornell University

This week: the arXiv Accessibility Forum

Forum Schedule
We gratefully acknowledge support from the Simons Foundation, member
institutions, and all contributors. Donate
 
arxiv logo > cs > arXiv:2408.16737
[                    ]

Help | Advanced Search

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

  * Login
  * Help Pages
  * About

Computer Science > Computation and Language

arXiv:2408.16737 (cs)
[Submitted on 29 Aug 2024]

Title:Smaller, Weaker, Yet Better: Training LLM Reasoners via
Compute-Optimal Sampling

Authors:Hritik Bansal, Arian Hosseini, Rishabh Agarwal, Vinh Q. Tran,
Mehran Kazemi
View a PDF of the paper titled Smaller, Weaker, Yet Better: Training
LLM Reasoners via Compute-Optimal Sampling, by Hritik Bansal and 4
other authors
View PDF HTML (experimental)

    Abstract:Training on high-quality synthetic data from strong
    language models (LMs) is a common strategy to improve the
    reasoning performance of LMs. In this work, we revisit whether
    this strategy is compute-optimal under a fixed inference budget
    (e.g., FLOPs). To do so, we investigate the trade-offs between
    generating synthetic data using a stronger but more expensive
    (SE) model versus a weaker but cheaper (WC) model. We evaluate
    the generated data across three key metrics: coverage, diversity,
    and false positive rate, and show that the data from WC models
    may have higher coverage and diversity, but also exhibit higher
    false positive rates. We then finetune LMs on data from SE and WC
    models in different settings: knowledge distillation,
    self-improvement, and a novel weak-to-strong improvement setup
    where a weaker LM teaches reasoning to a stronger LM. Our
    findings reveal that models finetuned on WC-generated data
    consistently outperform those trained on SE-generated data across
    multiple benchmarks and multiple choices of WC and SE models.
    These results challenge the prevailing practice of relying on SE
    models for synthetic data generation, suggesting that WC may be
    the compute-optimal approach for training advanced LM reasoners.

Subjects: Computation and Language (cs.CL); Artificial Intelligence
          (cs.AI)
Cite as:  arXiv:2408.16737 [cs.CL]
          (or arXiv:2408.16737v1 [cs.CL] for this version)
          https://doi.org/10.48550/arXiv.2408.16737
          Focus to learn more
          arXiv-issued DOI via DataCite

Submission history

From: Mehran Kazemi [view email]
[v1] Thu, 29 Aug 2024 17:32:35 UTC (233 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled Smaller, Weaker, Yet Better:
    Training LLM Reasoners via Compute-Optimal Sampling, by Hritik
    Bansal and 4 other authors
  * View PDF
  * HTML (experimental)
  * TeX Source
  * Other Formats

license icon view license
Current browse context:
cs.CL
< prev   |   next >
new | recent | 2024-08
Change to browse by:
cs
cs.AI

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?)
[ ] GotitPub Toggle
Gotit.pub (What is GotitPub?)
[ ] 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?)
[ ] Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
( ) 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