Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters
Snell et al., 2025, Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters (The Thirteenth International Conference on Learning Representations (ICLR 2025))
Subscriber deck — free during early access
2 cards in “5.3 Analysis Results: Test-Time Scaling for Search with Verifiers” · whole deck
5.3 Analysis Results: Test-Time Scaling for Search with Verifiers
- How did search methods compare as the budget grew?Beam search beat best-of-N at small budgets but its advantage faded and reversed at large budgets; lookahead search did worst for its cost. Stronger search over-optimized the PRM, producing repetitive low-information steps or very short solutions.
- Snell et al., 2025, Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters (The Thirteenth International Conference on Learning Representations (ICLR 2025)), §5.3
- How does the best search method depend on question difficulty?On easy questions beam search degraded with more budget (amplifying the verifier’s spurious features) while best-of-N didn’t; on medium-hard questions beam search consistently won; on the hardest, no method helped much.
- Snell et al., 2025, Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters (The Thirteenth International Conference on Learning Representations (ICLR 2025)), §5.3
Source
- Charlie Snell, Jaehoon Lee, Kelvin Xu and Aviral Kumar. 2025. Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters. In The Thirteenth International Conference on Learning Representations (ICLR 2025). https://arxiv.org/abs/2408.03314