T5 fine-tuned with LoRA on TweetSumm — running entirely in your browser
🌐 100% in-browser (WASM)🤖 T5-Small + LoRA r=4🗜️ INT8 · ~90 MB one-time download🔒 Your text never leaves this page
48
Loading model…
The summary will appear here. On the first run, the model
(~90 MB, INT8) is downloaded and cached by your browser — subsequent runs are instant.
💡 How it works
The LoRA adapter (rank 4, 0.24% of parameters) was merged into T5-Small,
exported to ONNX and quantized to INT8. Inference runs locally via
Transformers.js — no server, no API keys, no data sent anywhere.