Filtered by: VLLM × Clear all

ReSET: Accurate Latency-Critical NVFP4 Reasoning via Step-Aware Temperature Scaling

Sihwa Lee, Janghwan Lee, Donghoon Yoo, Jae Gon Kim 2026-06-14
Inference Quantization Hardware

Problem: Large reasoning models (LRMs) incur high inference costs due to long reasoning traces, and directly applying NVFP4 low-precision quantization degrades reasoning accuracy while existing kernels fail to deliver latency benefits in small-batch autoregressive decoding. Method: ReSET proposes a step-aware temperature scaling method that estimates step-level uncertainty online using both token-level and step-level entropy signals, and introduces a CUDA-core small-M NVFP4 kernel for latency-critical decoding. Finding: ReSET improves NVFP4 reasoning accuracy by up to ~2 points over the NVFP4 baseline, and the custom kernel achieves up to 2.5× kernel-level speedup over NVFP4 vLLM and approximately 2× end-to-end decoding speedup over BF16. Why it matters: This work enables accurate and efficient low-precision inference for latency-critical LRM deployments, reducing computational and memory costs without sacrificing reasoning quality.

PDF

GF-DiT: Scheduling Parallelism for Diffusion Transformer Serving

Xinwei Qiang, Yifan Hu, Shixuan Sun, Jing Yang 2026-06-14
Transformer GPU Runtime Scheduling Parallelism Serving

The problem is that existing Diffusion Transformer (DiT) serving systems use static parallelism for each request, which is inefficient due to heterogeneity across requests, execution stages, and system conditions. GF-DiT introduces a policy-programmable runtime that dynamically adapts parallelism via an asynchronous execution abstraction and group-free collectives for low-overhead online GPU reallocation. Experimental evaluation in vLLM-Omni shows GF-DiT improves throughput by up to 6.01×, reduces mean latency by up to 95%, and lowers SLO violation rates by up to 90% compared to fixed-pipeline execution. This matters because it enables efficient, elastic DiT serving that treats GPU parallelism as a schedulable resource, significantly improving performance and service quality for image and video generation workloads.

PDF