Tool / 08 · ONNX browser benchmark

EdgeBench

Run a local ONNX model with generated or JSON tensor inputs using WebGPU or WASM, then report p50, p95, sequential throughput, and a memory lower bound after explicit warm-up.