APISAvant2 / README.md
antonypamo's picture
Update README.md
6df19e8 verified
metadata
license: other
title: Savant RRF Φ12.0  Dirac-Resonant Conceptual Quality API
sdk: docker
emoji: 🐢
colorFrom: red
colorTo: green
pinned: true
short_description: API de evaluación conceptual resonante para LLM

🧠 Savant RRF Φ12.0 — Meta-Logic & Rerank API

Savant RRF Φ12.0 is a production-ready FastAPI service that exposes:

A meta-logic quality evaluator based on the Resonance of Reality Framework (RRF)

A batched semantic reranker using a custom icosahedral-resonant embedding model

A deterministic Φ-node ontology mapping layer

The system combines SentenceTransformer embeddings, spectral / resonance features, and a 15-dimensional meta-logit classifier to evaluate reasoning quality and rank documents efficiently.

🔗 Live API Base URL: https://antonypamo-apisavant2.hf.space

📦 Models Used Component Model Embedder antonypamo/RRFSAVANTMADE Meta-Logic antonypamo/RRFSavantMetaLogicV2/logreg_rrf_savant.joblib Feature Dim 15 features Runtime CPU (GPU optional if available) 🧩 Φ-Node Ontology

The system maps inputs to one of 8 deterministic Φ-nodes:

Index Φ Node 0 Φ0_seed 1 Φ1_geometric 2 Φ2_gauge_dirac 3 Φ3_log_gravity 4 Φ4_resonance 5 Φ5_memory_symbiosis 6 Φ6_alignment 7 Φ7_meta_agi

This mapping is rule-based and reproducible, derived from spectral coherence, energy, and phase features.

🚀 Endpoints Overview GET /

Root discovery endpoint.

{ "status": "ok", "model": "RRFSavantMetaLogicV2", "version": "Φ12.0", "docs": "/docs", "endpoints": ["/manifest", "/health", "/evaluate", "/quality", "/v1/rerank"] }

GET /health

Lightweight health check.

{ "status": "ok" }

GET /manifest

Static manifest describing the model.

{ "model": "RRFSavantMetaLogicV2", "version": "Φ12.0", "encoder": "antonypamo/RRFSAVANTMADE", "features": 15, "phi_nodes": [...] }

🧪 Quality Evaluation API POST /evaluate

Evaluates the conceptual quality of a (prompt, answer) pair.

Request { "prompt": "Explain what a smoke test is", "answer": "A smoke test is a minimal validation..." }

Response { "p_good": 0.87, "scores": { "SRRF": 0.87, "CRRF": 0.79, "E_phi": 0.82 }, "features": { "phi": 0.61, "omega": 0.22, "coherence": 0.83, "S_RRF": 0.81, "C_RRF": 0.77, "hamiltonian_energy": 48.3, "dominant_frequency": 12 }, "phi_node": "Φ4_resonance" }

POST /quality

Alias of /evaluate. Same input, same output.

🔍 Semantic Reranking API POST /v1/rerank

Ranks documents by semantic similarity to a query using batched embedding inference.

Request { "query": "What is a smoke test?", "documents": [ "A smoke test is a minimal system check", "Load tests measure concurrency", "Benchmarks compare systems" ], "alpha": 0.2 }

alpha is reserved for future hybrid scoring (currently unused).

Response { "model_id": "antonypamo/RRFSAVANTMADE", "results": [ { "id": 0, "score": 0.92, "rank": 1 }, { "id": 2, "score": 0.51, "rank": 2 }, { "id": 1, "score": 0.21, "rank": 3 } ] }

⚙️ Runtime Constraints Parameter Limit Max prompt length 8,000 chars Max answer length 12,000 chars Max documents 50 Max document size 6,000 chars

Payload violations return HTTP 413.

🧠 Feature Vector (15D)

The meta-logic classifier consumes:

Spectral / resonance features:

phi, omega, coherence, S_RRF, C_RRF

hamiltonian_energy, dominant_frequency

Φ-node one-hot encoding (8 dimensions)

Total = 15 features

🛠 Running Locally pip install fastapi uvicorn sentence-transformers huggingface_hub joblib numpy

export HF_TOKEN=your_token_here uvicorn app:app --host 0.0.0.0 --port 8000

Open:

http://127.0.0.1:8000/docs

📈 Performance Notes

Optimized for batched inference on /v1/rerank

Stable under load (0% error rate in benchmarks)

CPU-based by default; GPU reduces latency significantly

Tail latency (p95/p99) depends on concurrency and hardware

🧩 Design Philosophy

Savant RRF is not a generic classifier.

It encodes:

Discrete resonance physics

Icosahedral symbolic structure

Deterministic ontology mapping

Meta-logic scoring beyond surface semantics

This makes it suitable for:

AI evaluation & judging

RAG reranking

Cognitive profiling

Research-grade reasoning analysis

📄 License & Attribution

© 2025 Antony Padilla Morales Resonance of Reality Framework (RRF)