SOURCE-LINKED INTELLIGENCE

H-Scale: Hessian-Guided Scale Refinement for NVFP4 Sub-Byte LLM Inference

arXiv · AI, language, vision and robotics · article · Aug 28, 2026 · UTC

The NVIDIA Blackwell architecture, with native support for the ultra-fine-grained NVFP4 format, opens new opportunities for accelerating large language model (LLM) inference. NVFP4's micro-block design, such as a group size of 16, offers strong representational flexibility for capturing local weight distributions and isolating outliers, but it also introduces a large and highly sensitive space of per-group scaling factors. Existing post-training quantization (PTQ) methods primarily focus on refining quantized weight values, leaving this scale-selection step underexplored. To address this gap,

Read original source ↗ Open in workspace

recordType

paper

region

Global