ascend-benchmark-evaluator

What problem does it solve?

This Skill provides a structured framework to evaluate the Lingxi-Code Agent's Ascend C code generation across standardized datasets, producing both correctness verification and performance benchmarks.

Core Features & Use Cases

• Batch operator evaluation across multiple operators to generate per-operator reports.
• Correctness validation by comparing generated outputs against reference implementations to ensure functional parity.
• Performance benchmarking with timing comparisons and a consolidated benchmark report.
• Automated dataset handling (NPUKernelBench format) and end-to-end evaluation workflow.

Quick Start

Run the evaluator against your NPUKernelbench dataset to produce a complete per-operator and global benchmark report.