Personality Chat

Upload your trained model and chat with it entirely in your browser. No server needed — everything runs on your machine.

01·Train your model

Fine-tune SmolLM2-360M on your text using LoRA + Unsloth. Runs on your GPU or free Google Colab. Takes ~5-15 minutes.

02·Export to ONNX

The script automatically exports to ONNX with KV-cache support. The model files are bundled into a ZIP archive.

03·Upload & chat

Drag the ZIP into this page. The model loads into browser memory via ONNX Runtime Web. All inference runs locally — zero servers.

Upload your model ZIP

Contains model.onnx, config.json, tokenizer.json

Click to upload

Quick start

# Train + export on your GPU
python train/smol_lora_train.py \
  --data ./my-book.txt --steps 60 --export-onnx

# Upload the ZIP and start chatting

See it in action

Watch how the personality model runs entirely in your browser — no server involved