Hugging Face Papers ↗ · 2026-11-13
AscendGYM:面向 Ascend NPU 上真实工作负载的成本感知内核优化基准
Media reports Hugging Face Papers Source language: Chinese
论文摘要(作者报告,未经本站复测):内核优化对于在 Ascend NPU 上高效地进行大语言模型(LLM)训练和推理至关重要。
Full text
AscendGYM provides a benchmark for assessing kernel optimization capabilities on Ascend NPUs using real-workload test cases, two evaluation modes, and a Speedup-Token Score that balances speedup against token cost.