onnx-light — benchmark

Processing time comparison between onnxruntime and the onnx-light reference evaluator backed by its C++ kernel dispatch table (onnx_light.reference.ReferenceEvaluator) on every ONNX backend node test. Each test runs 3 warm-up iterations (not timed) and 10 timed iterations; the average time per dataset is reported. Speedup = onnxruntime avg / onnx‑light avg: values > 1 mean onnx‑light is faster.

⚠ The results displayed on this page are indicative and experimental.