AI Paper Atlas

Research map / Large language models / Alignment and preferences

PALC: Preference Alignment via Logit Calibration

SANGHYUN LEE, Hoh Peter In

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models

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 PALC logit calibration moduleLLaMA-7B-SFT · HH-RLHFwin+0.5tie rate vs base (%): 58.17 win+0.5tie; 54.67 winImproved
use PALC rather than static steering vectorsCAA · HH-RLHFwin+0.5tie rate: 77.17 win+0.5tie vs CAAImproved
use PALC rather than test-time optimizationRE-Control · HH-RLHFwin+0.5tie rate: 61.67 win+0.5tie vs RE-ControlImproved
use PALC rather than reward-model guided decodingARGS · HH-RLHFwin+0.5tie rate: 55.50 win+0.5tie vs ARGSNo clear effect
use PALC rather than activation steeringBiPO · HH-RLHFwin+0.5tie rate: 49.00 win+0.5tie vs BiPONo clear effect
use training-time DPO or GenARM instead of PALCPALC · HH-RLHFwin+0.5tie rate: 41.17 vs DPO; 44.33 vs GenARMHurt
vary bottleneck dimension from 16 to 1024 (ablation)PALC · HH-RLHFwin+0.5tie rate vs base (%): 53.7 to 58.2No clear effect
increase bottleneck dimension to 4096 (ablation)PALC · HH-RLHFwin+0.5tie rate vs base (%): 58.2 -> 18.3Hurt
scale calibration with gamma 0.5-3.0 (ablation)PALC · HH-RLHFwin+0.5tie rate vs base (%): 55.3 to 58.2No clear effect
raise scaling factor gamma to 10 (ablation)PALC · HH-RLHFwin+0.5tie rate vs base (%): 58.2 -> 38.7Hurt
use PALC rather than reward-model guided decoding for latencyGenARM · HH-RLHFrelative latency: 3.17x -> 1.08xImproved
apply PALC to MT-BenchLLaMA-7B-SFT · MT-Bench (LC win+tie vs base)length-controlled win+tie rate (%): 61.9 (GenARM 58.7)Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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