{"slug": "sample-count-is-not-enough-candidate-generation-strategy-shapes-the-energy-and", "title": "Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling", "summary": "A new arXiv paper (2609.19499v1) reports that generation schedule, not candidate count alone, determines the energy and latency cost of LLM test-time scaling: on A100 GPUs, eight serial generation calls with one candidate each used 4.64-4.86x as much gross GPU-device energy and had 5.77-6.12x the P95 latency of a single batched call with eight candidates at a fixed N = 8. The study, using Phi-3-mini and Qwen2.5-1.5B on 500 GSM8K prompts, found raising N from 1 to 8 improved accuracy by 8.4 percentage points for Phi-3-mini and 18.4 points for Qwen2.5-1.5B, with the same schedule pattern reproduced across three independently scheduled A100 nodes per model and in short-output SciQ/V100 experiments. The authors conclude evaluations should report generation schedule and GPU-level systems metrics alongside candidate count and accuracy, and that fewer calls with larger batches are more efficient when candidates are independent and memory allows.", "body_md": "arXiv:2609.19499v1 Announce Type: new \nAbstract: Test-time scaling can improve large language model reasoning by generating and combining multiple candidate responses. In sampling-based methods, the inference budget is often described by the number of generated candidates, N. However, N tells us how many candidates are generated, not how they are executed. The same candidate budget can be produced in one batched generation call or split across several sequential calls with smaller batch sizes. We first study the effect of increasing N on reasoning accuracy using Phi-3-mini and Qwen2.5-1.5B on 500 GSM8K prompts. As expected, increasing N from 1 to 8 improves accuracy by 8.4 percentage points for Phi-3-mini and 18.4 points for Qwen2.5-1.5B. However, accuracy alone does not show the systems cost of using a larger candidate budget. We therefore fix N = 8 and compare four generation schedules: 1x8, 2x4, 4x2, and 8x1, where axb denotes a generation calls with b candidates per call. We measure latency, throughput, GPU-hours, and gross GPU-device energy while keeping the total candidate count fixed. On A100 GPUs, eight serial calls use 4.64-4.86x as much gross GPU-device energy and have 5.77-6.12x the P95 latency of one batched call with eight candidates. The same pattern appears across three independently scheduled A100 nodes per model and in short-output SciQ/V100 experiments. These results show that candidate count alone is not enough to describe the systems cost of multi-candidate test-time scaling. When candidates are independent and memory allows it, fewer generation calls with larger batch sizes are more efficient. Evaluations should therefore report not only candidate count and accuracy, but also generation schedule and GPU-level systems metrics.", "url": "https://wpnews.pro/news/sample-count-is-not-enough-candidate-generation-strategy-shapes-the-energy-and", "canonical_source": "https://arxiv.org/abs/2609.19499", "published_at": "2026-09-18 04:00:00+00:00", "updated_at": "2026-09-18 04:23:53.373599+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-infrastructure", "mlops"], "entities": ["arXiv", "Phi-3-mini", "Qwen2.5-1.5B", "GSM8K", "SciQ", "A100", "V100"], "alternates": {"html": "https://wpnews.pro/news/sample-count-is-not-enough-candidate-generation-strategy-shapes-the-energy-and", "markdown": "https://wpnews.pro/news/sample-count-is-not-enough-candidate-generation-strategy-shapes-the-energy-and.md", "text": "https://wpnews.pro/news/sample-count-is-not-enough-candidate-generation-strategy-shapes-the-energy-and.txt", "jsonld": "https://wpnews.pro/news/sample-count-is-not-enough-candidate-generation-strategy-shapes-the-energy-and.jsonld"}}