Spaces:
Sleeping
Sleeping
| # OpenEnv manifest (Section 9.1 of the plan). | |
| # Validate with: openenv validate | |
| name: qubit-medic | |
| version: 1.0.0 | |
| description: | | |
| RL training environment for LLM-based quantum error-correction decoders. | |
| Built on Stim + PyMatching with five independent verifiable rewards | |
| (logical correction, syndrome consistency, Hamming overlap, format | |
| compliance, PyMatching beat-rate). Designed to reproduce the AlphaQubit | |
| (Nature 2024) two-stage decoder pipeline at distance-3 with off-the-shelf | |
| 3B-parameter LLMs trained on a single Colab T4. | |
| authors: | |
| - name: Qubit-Medic team | |
| license: MIT | |
| homepage: https://huggingface.co/spaces/qubit-medic/qubit-medic | |
| server: | |
| module: qubit_medic.server.app | |
| app: app | |
| protocol: http | |
| port: 7860 | |
| endpoints: | |
| # All endpoints below are registered automatically by | |
| # openenv.core.create_fastapi_app via qubit_medic.server.app. | |
| reset: | |
| method: POST | |
| path: /reset | |
| request_model: openenv.core.types.ResetRequest | |
| response_model: openenv.core.types.ResetResponse | |
| step: | |
| method: POST | |
| path: /step | |
| request_model: openenv.core.types.StepRequest | |
| response_model: openenv.core.types.StepResponse | |
| state: | |
| method: GET | |
| path: /state | |
| response_model: qubit_medic.server.openenv_adapter.QubitMedicState | |
| schema: | |
| method: GET | |
| path: /schema | |
| metadata: | |
| method: GET | |
| path: /metadata | |
| health: | |
| method: GET | |
| path: /health | |
| healthz: # Day-0 deployment-substrate probe (extra) | |
| method: GET | |
| path: /healthz | |
| decode: # PyMatching baseline demo (extra) | |
| method: POST | |
| path: /decode | |
| models: | |
| environment: qubit_medic.server.openenv_adapter.QubitMedicEnvironment | |
| action: qubit_medic.server.openenv_adapter.QubitMedicAction | |
| observation: qubit_medic.server.openenv_adapter.QubitMedicObservation | |
| state: qubit_medic.server.openenv_adapter.QubitMedicState | |
| rewards: | |
| - name: logical_correction | |
| weight: 0.40 | |
| description: 1 if the predicted Pauli frame preserves the logical Z observable. | |
| - name: syndrome_consistency | |
| weight: 0.20 | |
| description: Hamming similarity over final-round detector parities. | |
| - name: hamming_overlap | |
| weight: 0.20 | |
| description: Mean Jaccard similarity vs. PyMatching reference Pauli frame. | |
| - name: format_compliance | |
| weight: 0.10 | |
| description: 1 / 0.5 / 0 for full / partial / unparseable output. | |
| - name: pymatching_beat | |
| weight: 0.10 | |
| description: 1 iff PyMatching wrong AND model right on this syndrome. | |
| curriculum: | |
| - name: L1_warmup | |
| distance: 3 | |
| rounds: 1 | |
| p: 0.0001 | |
| promotion_threshold: 0.80 | |
| - name: L2_target | |
| distance: 3 | |
| rounds: 3 | |
| p: 0.001 | |
| promotion_threshold: 0.70 | |
| - name: L3_stretch | |
| distance: 5 | |
| rounds: 5 | |
| p: 0.001 | |
| promotion_threshold: 0.30 | |
| citations: | |
| - gidney_stim_2021: | |
| title: "Stim: a fast stabilizer circuit simulator" | |
| authors: Gidney, Craig | |
| venue: Quantum 5:497 | |
| doi: 10.22331/q-2021-07-06-497 | |
| arxiv: 2103.02202 | |
| note: | | |
| Field-standard Clifford simulator for quantum error correction. | |
| Same simulation substrate used by AlphaQubit (Bausch 2024) and | |
| Willow (Acharya 2024); not a homemade simulator. | |
| - bausch_alphaqubit_2024: | |
| title: Learning high-accuracy error decoding for quantum processors | |
| authors: Bausch et al. | |
| venue: Nature 635:834 | |
| doi: 10.1038/s41586-024-08148-8 | |
| - acharya_willow_2024: | |
| title: Quantum error correction below the surface code threshold | |
| authors: Acharya et al. (Google QAI) | |
| venue: arXiv:2408.13687 | |
| - gidney_si1000_2021: | |
| title: A fault-tolerant honeycomb memory | |
| authors: Gidney & Fowler | |
| venue: arXiv:2108.10457 | |
| - higgott_pymatching_2023: | |
| title: Sparse Blossom - PyMatching v2 | |
| authors: Higgott & Gidney | |
| venue: arXiv:2303.15933 | |
| - shao_grpo_2024: | |
| title: DeepSeekMath - introduces GRPO | |
| authors: Shao et al. | |
| venue: arXiv:2402.03300 | |