| --- |
| library_name: pytorch |
| tags: |
| - quantum-error-correction |
| - circuit-family-classifier |
| license: other |
| license_name: nvidia-open-model-license-derivative |
| --- |
| |
| # QFabric Shield — Family Classifier (per-distance) |
|
|
| Lightweight transformer classifiers that ingest surface-code syndrome history |
| and emit a softmax over four QFabric circuit families: |
| `{mps, grover, tebd, emu}`. Used at runtime by the Shield dispatcher to route |
| syndromes to the correct family-specialized decoder. |
|
|
| One classifier per code distance (input dimension scales with D). |
|
|
| ## Architecture |
|
|
| - 4-layer transformer encoder, d_model=128, 4 heads |
| - ~1M parameters |
| - Input: `(B, 4, T, D, D)` syndrome tensor |
| - Output: `(B, 4)` family logits |
| |
| ## Distance variants |
| |
| Each `d{{N}}/` subdir contains the classifier trained at that distance. |
| |
