Compute Aligned Training: Optimizing for Test Time Inference

TL;DR

This paper introduces Compute Aligned Training, a method that aligns training objectives with test-time inference strategies to improve Large Language Model performance at scale.

Contribution

It proposes new loss functions that optimize training for specific test-time inference strategies, bridging the gap between training and inference procedures.

Findings

Significant improvement in test time scaling performance.

Effective alignment of training objectives with inference strategies.

Empirical validation across supervised fine-tuning and reinforcement learning.

Abstract

Scaling test-time compute has emerged as a powerful mechanism for enhancing Large Language Model (LLM) performance. However, standard post-training paradigms, Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), optimize the likelihood of individual samples under a base policy, creating a misalignment with test time procedures that rely on aggregated or filtered outputs. In this work, we propose Compute Aligned Training, which aligns training objectives with test-time strategies. By conceptualizing inference strategies as operators on the base policy, we derive new loss functions that maximize performance when said strategies are applied. We instantiate such loss functions for SFT and RL across common test time strategies. Finally, we provide empirical evidence that this training method substantially improves test time scaling over standard training.