Beyond Correctness: Compute-Responsive Data Selection for Reasoning Model Post-Training

Key points

• We ask a complementary question: can reasoning models be post-trained to make better use of a given amount of test-time compute, and which examples are most useful for this goal?1
• Our key observation is that reasoning problems differ substantially in their response to additional computation.1
• We call these compute-responsive (CR) examples and introduce Compute-Responsive Data Selection (CRDS), a post-training data-selection principle that prioritizes examples with empirically demonstrated marginal utility from additional reasoning.1
• CR training also shifts the compute--accuracy frontier across inference budgets from 512 to 4,096 tokens, outperforming matched non-CR training throughout and reaching 48.1% recovery at the largest budget.1

Sources (1)

• [1]Beyond Correctness: Compute-Responsive Data Selection for Reasoning Model Post-Training
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 11:57 AM

  Reasoning language models increasingly rely on test-time compute scaling, yet most existing work studies how to allocate inference computation while treating the model itself as fixed.

  We ask a complementary question: can reasoning models be post-trained to make better use of a given amount of test-time compute, and which examples are most useful for this goal?

Extractive summary: sentences quoted from the sources.