Search papers, labs, and topics across Lattice.
The paper introduces Compute Aligned Training (CAT), a novel training paradigm that aligns training objectives with test-time inference strategies used in LLMs. CAT conceptualizes inference strategies as operators on the base policy and derives new loss functions for SFT and RL that maximize performance when these strategies are applied. Experiments demonstrate that CAT significantly improves test-time scaling compared to standard training methods.
Training LLMs to explicitly optimize for how they're *actually* used at inference time unlocks substantial performance gains compared to standard fine-tuning.
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.
