GPU-NFBench: A Human-Labeled Benchmark and LLM-Assisted Triage System for GPU Numerical Failures
收藏官方服务：
资源简介：
GPU-NFBench is a 1,191-row human-labeled benchmark and reproducible artifact packet for classifying GPU numerical failure reports from public GitHub issues. The artifact includes v2 adjudicated labels, deterministic baselines, fine-tuned LLM evaluation outputs, selective triage reports, root-cause evidence files, annotation audit files, and the IEEE-format manuscript.
提供机构：
Zenodo创建时间：
2026-05-16
