AI Paper Atlas

Research map / Large language models / Alignment and preferences

LLMdoctor: Token-Level Flow-Guided Preference Optimization for Efficient Test-Time Alignment of Large Language Models

Tiesunlong Shen, Rui Mao, Jin Wang, Heming Sun, Jian Zhang, Xuejie Zhang, Erik Cambria

AAAI 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 5 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 subtrajectory balance loss (ablation)LLMdoctor · HH-RLHF testWin + 1/2 Tie (%) vs. DPO: 53.15 -> 61.00Improved
add subtrajectory balance loss (ablation)LLMdoctor · HH-RLHF testDiversity: 0.34 -> 0.47Improved
add value discrimination loss (ablation)LLMdoctor · HH-RLHF testWin + 1/2 Tie (%) vs. DPO: 58.23 -> 61.00Improved
add value discrimination loss (ablation)LLMdoctor · HH-RLHF testDiversity: 0.43 -> 0.47Improved
add reward sparsity threshold (ablation)LLMdoctor · HH-RLHF testWin + 1/2 Tie (%) vs. DPO: 56.58 -> 61.00Improved
add reward sparsity threshold (ablation)LLMdoctor · HH-RLHF testDiversity: 0.46 -> 0.47No clear effect
add flow-guided rewards (ablation)LLMdoctor · HH-RLHF testWin + 1/2 Tie (%) vs. DPO: 52.76 -> 61.00Improved
add flow-guided rewards (ablation)LLMdoctor · HH-RLHF testDiversity: 0.25 -> 0.47Improved
use LLMdoctor instead of GenARM as test-time guidanceLLaMA-7B-SFT · HH-RLHF testWin + 1/2 Tie (%): 62.10Improved
use LLMdoctor test-time guidance instead of DPO trainingLLaMA-7B-SFT · HH-RLHF testWin + 1/2 Tie (%): 61.00Improved
use 7B doctor to guide 70B patient instead of DPO fine-tuningTulu2-SFT 70B · AlpacaEval 2LC win rate (%): 82.0 -> 82.5Improved
use ARGS instead of DPOLLaMA-7B-SFT · HH-RLHF testWin + 1/2 Tie (%): 26.24Hurt

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