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

Home/Benchmarks

Bias QA on BBQ Disambig

84.85Accuracy

CRL-Token

Updated 1mo ago

Evaluation Results

Method	Links
CRL-Token 2026.02		84.85	0.47
Gemma 2 2B 2026.02		84.38	-

SOTA Paper

CRL-Token

Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features

Dataset

BBQ

Follow for update

@wizwand_team Discord

Related Benchmarks

Bias QA on BBQ Ambig Short-form QA on SimpleQA Math Reasoning on GSM8K (Accuracy and Gain)Adversarial Safety on HarmBench Multiple Choice Question Answering on MMLU (Performance Change Tracking)

© 2026 wizwand

Blog Contact Changelog Swarm

Privacy Policy Terms of Service FAQs Swarm Docs