Firehose

16 SEP 2026 · Paper

Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling

This paper investigates how the way large language models generate multiple candidate responses affects their performance and energy consumption. Practitioners might care because optimizing test-time scaling can lead to significant improvements in model accuracy and efficiency.

large language models test-time scaling continual learning energy efficiency GPU-level systems metrics

Browse by tag

Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling