vela-1.0-encoder-307m-embedding__vela_l128_embeds_fp16sentence-similarity · family: vela-1.0-encoder-307m-embedding
| Source | https://huggingface.co/litert-community/Vela-1.0-Encoder-307M-Embedding-LiteRT |
|---|---|
| License | apache-2.0 |
| Model file | vela_L128_embeds_wfp16_shift_arith.tflite — downloaded from the source at runtime; this site hosts no weights |
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.
| Backend | Status | Full delegation | Output match | Latency p50 (ms) | Environment | Verified | Provenance |
|---|---|---|---|---|---|---|---|
wasm_xnnpack | pass | — | — | 399.72 | mac-studio-m4-max · chromium 151.0.7922.34 headless · macOS · @litertjs/core 2.5.3 | 2026-10-18 stale — latest known 2.6.0 | measured |
webgpu_mldrift | pass | yes | pass | 19.13 | mac-studio-m4-max · chromium 151.0.7922.34 headless · macOS · @litertjs/core 2.5.3 | 2026-10-18 stale — latest known 2.6.0 | measured |
Full record (failure class, evidence, sweep config): raw sweep JSON · card.json · CARD.md
One-line prompt for your AI coding assistant (uses the litert-cookbook web skill):
Use the litert-cookbook web skill to build a LiteRT.js browser demo for vela-1.0-encoder-307m-embedding__vela_l128_embeds_fp16: load the model from https://huggingface.co/litert-community/Vela-1.0-Encoder-307M-Embedding-LiteRT/resolve/main/vela_L128_embeds_wfp16_shift_arith.tflite (license: apache-2.0) and run it with @litertjs/core.
One-line prompt for Claude Code / Gemini CLI — re-runs the same sweep verification for this model, locally:
In the edge-compat repo, verify vela-1.0-encoder-307m-embedding__vela_l128_embeds_fp16 with the sweep harness: write /tmp/catalog.csv with the header line model_id,source,input_spec,license,notes and the single row vela-1.0-encoder-307m-embedding__vela_l128_embeds_fp16,https://huggingface.co/litert-community/Vela-1.0-Encoder-307M-Embedding-LiteRT/resolve/main/vela_L128_embeds_wfp16_shift_arith.tflite,1x128x768:float32;1x128:float32,apache-2.0, — then run: cd web/sweep && npm ci && npm run sweep -- --catalog /tmp/catalog.csv --out out/ — and compare out/vela-1.0-encoder-307m-embedding__vela_l128_embeds_fp16.json with the sweep table on this page.