AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

FairJudge : An Adaptive, Debiased, and Consistent LLM-as-a-Judge

Bo Yang, Lanfei Feng, Yunkui Chen, Xiao Xu, Yu Zhang, Shijian Li

ICML 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add SFT stage (rubric-aware supervised fine-tuning) (ablation)FairJudge-8B · PandaLM Test Setmacro-F1: 68.74 -> 72.18Improved
add SFT stage (ablation)FairJudge-8B · FairJudge-Benchmark-1Kmacro-F1: 57.86 -> 67.63Improved
add SFT stage (ablation)FairJudge-8B · JudgeLM Test Setmacro-F1: 63.86 -> 66.78Improved
add DPO debiasing stage (ablation)FairJudge-8B · PandaLM Test Setmacro-F1: 70.05 -> 72.18Improved
add DPO debiasing stage (ablation)FairJudge-8B · JudgeLM Test SetRecall: 67.17 -> 66.93Hurt
add GRPO consistency stage (ablation)FairJudge-8B · FairJudge-Benchmark-1KPrecision: 72.32 -> 71.15Hurt
add GRPO consistency stage (ablation)FairJudge-8B · JudgeLM Test Setmacro-F1: 63.95 -> 66.78Improved
full SFT-DPO-GRPO training of Qwen3-VL-8B (ablation)Qwen3-VL-8B · PandaLM Test Setmacro-F1: 54.05 -> 72.18Improved
add cross-mode consistency training (GRPO) and policy framingFairJudge-8B · Pointwise-pairwise consistency, FairJudge-Benchmark-1Kconsistency (%): 60.59 -> 65.52Improved
scale general-purpose model to 72B or 671B instead of judge traininggeneral-purpose LLM judges · PandaLM, JudgeLM, FairJudge-Benchmark-1Kagreement/F1Mixed
increase FairJudge size from 2B to 4B to 8BFairJudge · PandaLM Test Setmacro-F1: 61.68 -> 62.18 -> 72.18Mixed
use Fast mode (final decision only) instead of Full reasoningFairJudge-8B · 1,000 samples on RTX 4090wall-clock time: 16 min 52 s -> 1 min 19 sImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy