Share your thoughts, 1 month free Claude Pro on usSee more

Home/Benchmarks

Closed Loop RLHF on HH 0% noise

61.6Win Rate

α-URM

Updated 19d ago

Evaluation Results

Method	Links
α-URM 2026.07		61.6
URM 2026.07		60.3
RM 2026.07		55

SOTA Paper

α-URM

Unbiased Alignment for Large Language Models with Noisy Preferences

Dataset

HH

Follow for update

@wizwand_team Discord

Related Benchmarks

Dialogue Preference Learning on HH (test)Summarization Preference Learning on TL;DR (test)Dialogue on HH (Anthropic Helpful and Harmless)Summarization on TL;DR (Win Rate variants)Reward Model Learning on UFB

© 2026 wizwand

Blog Contact Changelog Swarm

Privacy Policy Terms of Service FAQs Swarm Docs