AI WebGPU Lab Benchmark

Fixed Scenario Runtime Shootout

`bench-runtime-shootout` is now a repo-specific benchmark harness. It runs the same prompt and output budget across three deterministic browser-side runtime profiles and produces a fixed comparison surface.

Use this harness to freeze the benchmark protocol first. Replace the synthetic profiles with real runtimes only after install/build/deploy paths are decided.

Benchmark Controls

Run the full benchmark to compare all profiles under the same prompt, context, and output budget.

Comparison Notes

  • Prompt and output token budget stay fixed across all profiles.
  • Profiles differ only in synthetic runtime shape: init delay, prefill chunking, decode chunking, and worker mode.
  • The best profile is chosen from the combined TTFT and decode throughput score for the benchmark draft.

Last Comparison Matrix

No benchmark run yet.

Metrics

Environment

Activity Log

    Schema-Aligned Result Draft

    {
  "status": "pending"
}
    Open Repository Read README View Results