Live · Runs in your browser
MNIST CNN
A CNN I trained in PyTorch reads a handwritten digit as you draw it.
Draw a digit from 0 to 9 on the pad.
Loading model
Prediction-
- 00.0%
- 10.0%
- 20.0%
- 30.0%
- 40.0%
- 50.0%
- 60.0%
- 70.0%
- 80.0%
- 90.0%
PyTorch CNN exported to ONNX, running in your browser.
ONNX Runtime Web (WASM) · 1.7 MB model · 0 bytes uploaded
Under the hood
- Model
- 2 conv blocks + dense head
- Weights
- 421K parameters
- Input
- 1 × 1 × 28 × 28, normalised
- Output
- 10 class logits, softmaxed
- Training
- PyTorch, Adam, 10 epochs on MNIST
- Runtime
- ONNX Runtime Web, WASM