AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Towards Acyclic Preference Evaluation of Language Models via Multiple Evaluators

Zhengyu Hu, Jieyu Zhang, Zhihan Xiong, Alexander Ratner, Kaize Ding, Ranjay Krishna

AAAI 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems · 17 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 graph denoising to ensemble (ablation)PGED · response selection, HumanEvalaccuracy: 69.25 -> 71.86Improved
add graph denoising to ensemble (ablation)PGED · response selection, average of 5 benchmarksaverage score: 54.87 -> 56.98Improved
add graph denoising to single evaluator Qwen2-72B (ablation)single evaluator · response selection, HumanEvalaccuracy: 60.97 -> 68.90Improved
add graph denoising to single evaluator Llama3-8B (ablation)single evaluator · response selection, average of 5 benchmarksaverage score: 52.04 -> 53.33Improved
ensemble preference graphs before denoising instead of denoising then ranking ensemble (ablation)PGED · model ranking, AlpacaEval 2.0Spearman avg: 67.27 -> 72.50Improved
add graph denoising to ensemble (ablation)PGED · model ranking, AlpacaEval 2.0Spearman avg: 70.72 -> 72.50Improved
add graph denoising to single evaluator Qwen2-72B (ablation)single evaluator · model ranking, AlpacaEval 2.0Spearman avg: 65.04 -> 70.32Improved
combine small evaluators with PGED instead of using Qwen2-72B aloneevaluators · response selection, HumanEvalaccuracy: 60.97 -> 71.86Improved
use PGED instead of Majority Votingmultiple evaluators · response selection, average of 5 benchmarksaverage score: 53.73 -> 56.98Improved
list all candidates in one prompt instead of pairwise comparison (ListPreference)Qwen2-72B · response selection, average of 5 benchmarksaverage score: 53.07 -> 52.16Hurt
use PGED data selection instead of random selectionLlama-2-7B · AlpacaEval 2.0 win ratewin rate: 47.10% -> 53.50%Improved
use PGED data selection instead of ContraSolverLlama-2-7B · Koala win ratewin rate: 63.36% -> 65.91%Improved

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