Pith. sign in

REVIEW 5 cited by

InfAlign: Inference-aware language model alignment

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2412.19792 v5 pith:AZU3KX4S submitted 2024-12-27 cs.LG cs.CLcs.ITmath.IT

InfAlign: Inference-aware language model alignment

classification cs.LG cs.CLcs.ITmath.IT
keywords inference-timemodelrewardalignmentlanguagestandardalignedbest-of-n
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Language model alignment is a critical step in training modern generative language models. Alignment targets to improve win rate of a sample from the aligned model against the base model. Today, we are increasingly using inference-time algorithms (e.g., Best-of-N, controlled decoding, tree search) to decode from language models rather than standard sampling. We show that this train/test mismatch makes standard RLHF framework sub-optimal in view of such inference-time methods. To this end, we propose a framework for inference-aware alignment (InfAlign), which aims to optimize inference-time win rate of the aligned policy against the base model. We prove that for any inference-time decoding procedure, the optimal aligned policy is the solution to the standard RLHF problem with a transformation of the reward. This motivates us to provide the calibrate-and-transform RL (InfAlign-CTRL) algorithm to solve this problem, which involves a reward calibration step and a KL-regularized reward maximization step with a transformation of the calibrated reward. For best-of-N sampling and best-of-N jailbreaking, we propose specific transformations offering up to 3-8% improvement on inference-time win rates. Finally, we also show that our proposed reward calibration method is a strong baseline for optimizing standard win rate.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. What should post-training optimize? A test-time scaling law perspective

    cs.LG 2026-05 unverdicted novelty 6.0

    Tail-extrapolated estimators approximate best-of-N policy gradients from limited training rollouts by leveraging upper-tail reward statistics under structural assumptions.

  2. CLR-voyance: Reinforcing Open-Ended Reasoning for Inpatient Clinical Decision Support with Outcome-Aware Rubrics

    cs.CL 2026-05 unverdicted novelty 6.0

    CLR-voyance reformulates inpatient reasoning as POMDP with clinician-validated outcome rubrics, yielding an 8B model that outperforms larger frontier models on the authors' new benchmark.

  3. When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient

    cs.LG 2026-04 unverdicted novelty 6.0

    Certain errors in proxy rewards for policy gradient methods can be benign or beneficial by preventing policies from stalling on outputs with mediocre ground truth rewards, enabling improved RLHF metrics and reward des...

  4. Representation-Based Exploration for Language Models: From Test-Time to Post-Training

    cs.LG 2025-10 conditional novelty 6.0

    Representation-based elliptical bonuses improve inference-time and post-training pass@k for LLM reasoning, but the headline AIME result is tainted by validation/test overlap.

  5. Inference-Time Policy Alignment for Fair Reinforcement Learning

    cs.LG 2026-07 reject novelty 5.0

    A frozen RL policy can be reweighted at test time by a learned generalized-Gini welfare critic to improve fairness metrics, though the central equivalence mixes up two different welfare objectives.