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
1 cards in “2 A Unified Perspective on Test-Time Computation: Proposer and Verifier” · whole deck
2 A Unified Perspective on Test-Time Computation: Proposer and Verifier
- What two “knobs” does Snell et al.’s framework use to improve outputs at test time?Modifying the proposal distribution (e.g. a model fine-tuned to revise its own previous answers) and searching against a verifier (e.g. best-of-N or tree search scored by a process reward model), analogous to MCMC combining a proposal distribution with a score function.
- 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)), §2
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