AI WebGPU Lab Experiment

Browser Reranker Readiness

`exp-reranker-browser` now exposes a deterministic browser reranker harness. It reranks a fixed candidate set for one query, records scoring latency, top-k quality, and fallback-ready backend metadata before a real model runtime is connected.

Query

Run Controls

Run the reranker to score fixed candidates, sort top-k output, and export schema-aligned latency and quality metrics.

Ranking Output

No reranker run yet.

Metrics

Environment

Activity Log

    Schema-Aligned Result Draft

    {
  "status": "pending"
}

    Readiness Notes

    • Candidate set, expected best document, and query budget stay fixed.
    • Embedding metric fields carry reranker throughput, p50/p95 latency, and top-k quality until a dedicated reranker metric schema is added.
    • The harness keeps the same result contract for later Transformers.js, ORT-Web, or WebGPU runtime replacement.