AscendGYM: a cost-aware benchmark of 48 kernels for optimizing Ascend NPU kernels
SLAIAITP · hf · 2026-11-18
The AI Training Platform team at Shenzhen Loop Area Institute released AscendGYM, a benchmark for kernel optimization on Ascend NPUs targeting real workloads rather than just kernel generation.
- 48 kernels across three difficulty levels
- Test cases use data types and tensor shapes collected from real training and inference tasks, so gains map directly to workloads
- Supports model evaluation and model-harness evaluation modes for both trained models and custom agents
- Introduces a Speedup-Token Score balancing speedup against token cost
More from Infra
- Stratechery: Nvidia's earnings are a statement; Microsoft and Anthropic ship dueling AI OS updates — Stratechery · 2026-11-18
- Codeberg goes dark: domain vanished, pages unconfigured — jedisct1 · 2026-11-18
- All hyperscalers but Microsoft post negative FCF as AI capex peaks — marlene_zw · 2026-11-18
- Codeberg's domain mysteriously vanishes from its registrar account, site down — kingslyj · 2026-11-18
- Tokkey debuts thumb-sized AI hardware to run local models on your Mac — TejasKumar_ · 2026-11-18
- Hugging Face ships caching and serving upgrades, TTFT 1.5x faster — osanseviero · 2026-11-18