blueprint-ai/symsts-minilm-demo

SymSTS-MiniLM · Sentence Similarity — running 100% in your browser

A free static Space that runs [blueprint-ai/SymSTS-MiniLM](https://huggingface.co/blueprint-ai/SymSTS-MiniLM) entirely client-side via ONNX Runtime Web (WASM) + the model exported to fp16 ONNX and mean-pooled exactly like the original sentence-transformers pipeline.

• —No server, no GPU, no PRO tier — inference happens on your own CPU in the browser.
• —Private — your text never leaves the device.
• —Scores are numerically verified against the PyTorch model (identical to <0.01).

Files:

• —model.onnx — SymSTS-MiniLM exported to fp16 ONNX (45 MB)
• —vocab.txt — BERT WordPiece vocab
• —mybert.js — minimal BERT tokenizer + mean-pool (validated byte-exact against transformers)
• —index.html / app.js — UI + onnxruntime-web wiring