{"slug": "ascendgym-a-cost-aware-kernel-optimization-benchmark-for-real-world-workloads-on", "title": "AscendGYM: A Cost-Aware Kernel Optimization Benchmark for Real-World Workloads on Ascend NPUs", "summary": "Researchers released AscendGYM, a benchmark for evaluating how well AI models and agents optimize existing kernels on Ascend NPUs, with an initial suite of 48 kernels across three difficulty levels. AscendGYM pairs each kernel with test cases whose data types and tensor shapes come from real-world training and inference tasks, supports model and model-harness evaluation modes, and introduces the Speedup-Token Score (STS) metric to measure performance gains relative to token cost. The benchmark is available at https://github.com/SLAI-AITP/AscendGYM.", "body_md": "arXiv:2611.17621v1 Announce Type: new \nAbstract: Kernel optimization is essential for efficient large language model (LLM) training and inference on Ascend NPUs. Existing Ascend kernel benchmarks primarily focus on kernel generation, with less emphasis on the ability of agents to optimize existing implementations under input configurations drawn from real workloads. To address this requirement, we introduce AscendGYM, a benchmark whose initial suite comprises 48 kernels across three difficulty levels. Each kernel is paired with test cases whose data types and tensor shapes are collected from real-world training and inference tasks. Consequently, performance improvements on AscendGYM map directly to the corresponding workloads. In terms of usability, two evaluation modes are supported in AscendGYM: model evaluation and model-harness evaluation. The kernel optimization capabilities of both trained models and custom agents are allowed to be measured. It provides a complete execution environment that closely reproduces how Ascend kernels are invoked in practical pipelines. For model evaluation, it additionally includes a built-in harness equipped with search orchestration and experience accumulation. Finally, we propose a new metric, the Speedup-Token Score (STS), which quantifies performance gains relative to token cost. By jointly considering speedup and cost, STS enables the assessment of practical kernel optimization capability of a model or harness under an acceptable budget. AscendGYM is available at https://github.com/SLAI-AITP/AscendGYM.", "url": "https://wpnews.pro/news/ascendgym-a-cost-aware-kernel-optimization-benchmark-for-real-world-workloads-on", "canonical_source": "https://arxiv.org/abs/2611.17621", "published_at": "2026-11-16 05:00:00+00:00", "updated_at": "2026-11-16 05:20:12.873237+00:00", "lang": "en", "topics": ["ai-research", "machine-learning", "ai-infrastructure", "ai-agents", "ai-chips"], "entities": ["AscendGYM", "Ascend NPUs", "Speedup-Token Score", "SLAI-AITP"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/ascendgym-a-cost-aware-kernel-optimization-benchmark-for-real-world-workloads-on", "markdown": "https://wpnews.pro/news/ascendgym-a-cost-aware-kernel-optimization-benchmark-for-real-world-workloads-on.md", "text": "https://wpnews.pro/news/ascendgym-a-cost-aware-kernel-optimization-benchmark-for-real-world-workloads-on.txt", "jsonld": "https://wpnews.pro/news/ascendgym-a-cost-aware-kernel-optimization-benchmark-for-real-world-workloads-on.jsonld"}}