๐ฎ 1000-Qubit Quantum MNIST Classifier
Pure quantum ML model โ trained entirely on QPU-1 with zero classical neural network.
Architecture
| Component | Qubits | Details |
|---|---|---|
| Input | 0-783 | Ry angle encoding (28ร28 pixels) |
| Variational | 784-989 | 206 qubits, 2 layers (Ry + CNOT) |
| Output | 990-999 | 10 qubits (one per digit class) |
| Total | 1000 |
- 422 trainable parameters (Ry rotation angles)
- Training method: Parameter-shift rule (fully quantum gradients)
- Compute: QPU-1 by Lap Quantum
How it works
- Encoding: Each MNIST pixel is encoded as a Ry rotation on input qubits
- Variational layers: Parameterized Ry gates + CNOT entanglement ladders
- Cross-entanglement: Input qubits connected to variational qubits
- Readout: 10 output qubits measured; argmax = predicted digit
- Gradients: Parameter-shift rule โ shift each angle by ยฑฯ/2, measure loss difference
Train it yourself
Use the trainer Space: Reality123b/quantum-mnist-1000qubit-trainer
Awaiting first training run โ click "Train on QPU-1" in the Space to generate weights.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support