Does your .tflite run in the browser? Drop it in and find out in seconds — it loads, compiles, and runs right here, on your hardware.
Drop a .tflite file here — or pick one:
Preparing…
Every number is labeled with the environment that produced it — headless desktop Chromium is not a user’s phone, and results are only meaningful together with their environment. The numbers below were measured just now in your environment (shown underneath) — they are not the reference measurements from the zoo table.
| Backend | Loads | Runs | Full delegation | Output match | Latency p50 (ms) | Failure |
|---|
The operator inventory is parsed from your model locally (operator names only — full graph analysis is what edge-lint does). Statuses come only from the exported matrix snapshots listed below — real measurements, never inferred by this page. Ops without a matrix entry read unknown, and honestly so: op-level browser data exists only where a runtime log named the op.
| Backend | Verified against @litertjs/core | Generated | Entries | Snapshot |
|---|---|---|---|---|
wasm_xnnpack | 2.5.3 | 2026-08-12 | 127 | raw JSON |
webgpu_mldrift | 2.5.3 | 2026-08-12 | 132 | raw JSON |
Full partition analysis (partition map, blocking-op ranking, rewrite hints) runs locally with edge-lint — your model stays on your machine there too:
One-line prompt for your AI coding assistant (uses the litert-cookbook web skill):
One-line prompt for Claude Code / Gemini CLI — the same static verdicts locally with full partition analysis (your model stays on your machine):
In the edge-compat repo, reproduce this page's static verdicts for my model locally — the model stays on my machine: uv sync && uv run edge-lint MODEL.tflite --matrix data/matrix/wasm_xnnpack__2.5.3.json --json.