ReSET: Temperature Scaling Method Improves Accuracy of Low-Precision Reasoning Models
Researchers have proposed ReSET, a technique that uses entropy-based temperature scaling to recover reasoning accuracy lost when large language models are run under NVFP4 low-precision quantization. NVFP4 quantization reduces computational and memory costs but degrades model accuracy by mishandling token-level uncertainty at both low- and high-entropy reasoning steps. ReSET, combined with a custom CUDA kernel, restores up to ~2 accuracy points and delivers up to 2.5× kernel-level and ~2× end-to-end decoding speedups, making efficient deployment of reasoning models more practical.
Large reasoning models (LRMs) generate extended chains of intermediate reasoning steps to solve complex problems, but this significantly raises inference costs in both compute and memory. NVFP4 quantization is a hardware-supported low-precision format that can reduce these costs, yet applying it directly to LRMs causes accuracy degradation and fails to fully realize latency benefits in small-batch autoregressive decoding. The authors analyzed how NVFP4 quantization distorts token-level uncertainty, finding that it increases incorrect sampling at low-entropy symbolic tokens while over-concentrating probability mass at high-uncertainty reasoning steps. To address this, they developed ReSET (Reasoning-Step Entropy-based Temperature Scaling), which estimates step-level uncertainty online and dynamically adjusts decoding temperature using both token-level and step-level entropy signals. Separately, they designed a CUDA-core small-M NVFP4 kernel optimized for the latency-critical small-batch decoding regime. Across multiple reasoning benchmarks and model scales, ReSET recovers up to approximately 2 accuracy points over the NVFP4 baseline, while the new kernel achieves up to 2.5× kernel-level speedup over NVFP4 vLLM and roughly 2× end-to-end speedup over BF16 decoding. Code has been made publicly available.
What's missing
The accuracy recovery of ~2 points is described as 'up to,' and average-case gains across all benchmarks and model scales are not clearly summarized. It is also unclear how ReSET interacts with other quantization formats (e.g., INT8, FP8) or whether the custom CUDA kernel is compatible with inference frameworks beyond vLLM. The work is a preprint and has not yet undergone peer review.
What different sources said
- arXiv cs.AICenter
ReSET: Accurate Latency-Critical NVFP4 Reasoning via Step-Aware Temperature Scaling
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.