Benchmark RadarAI BENCHMARK PROFILE
Deep4ge
Deep4ge is a dataset of 14,227 training runs from 59 DNN programs with documented faults, providing per-epoch features for fault detection and diagnosis tasks, including binary detection, multi-class diagnosis, and early prediction.
- Released
- 2026-07-14
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
This fills the gap of a public dataset for diagnosing DNN implementation faults, supporting reproducible research and tool development for fault detection in deep learning systems.
Motivation
Deep learning systems often fail due to subtle implementation faults that alter training behavior.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.