kofdai/AXIS-Sovereign-Logic-Engine

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:2.6BQuant:BF16Ctx Length:8kPublished:Dec 28, 2025License:apache-2.0Architecture:Transformer Open Weights Warm

AXIS-Sovereign-Logic-Engine by kofdai is an Advanced Cross-Integrated System (AXIS) designed for deterministic operations and intelligence sovereignty. It employs a unique 'Lathe' architecture where an AI generates logical components, which are then rigorously validated by an external 'Verifier' using a rejection protocol. This system aims to eliminate hallucination by physically purging failed reasoning from memory and assembling final responses from verified raw data using templates, making it suitable for applications requiring high logical consistency and factual accuracy.

Loading preview...

AXIS: Advanced Cross-Integrated System (V1.6)

AXIS is a novel architecture that treats AI as a non-deterministic generator (a "Lathe") and pairs it with a deterministic "Verifier" to ensure complete control over outputs. This system is engineered to establish intelligence sovereignty and achieve deterministic computation.

Core Engineering Principles

  • Lathe Architecture & Rejection Protocol: The AI unit generates logical components, which are then subjected to a rigorous rejection loop. An external verifier (e.g., Python/SymPy) checks AI-proposed solutions against constraints. Any contradiction leads to immediate rejection, a session reset, and forced regeneration.
  • Context Purge: To prevent hallucination chains (Context Drift), torch.mps.empty_cache() is used to physically clear the "failed reasoning" from memory, ensuring statistical independence between trials.
  • 5D Semantic Lattice: Nodes are managed as SemanticNode classes, utilizing five orthogonal parameters for precise coordinate management:
    • s1: Physical Actuality (numerical/constant consistency)
    • s2: Logical Necessity (derivability from axioms)
    • s3: Contextual Dependency (consistency with Context Stack)
    • s4: Ethical Score (adherence to safety protocols)
    • s5: Empirical History (match count with past confirmed data)
  • Persistence & Acceleration: A "Semantic ID" (a unique hash from vector quantization) is used as a key for a high-density cache (local_massive_data.json). This allows the system to bypass AI inference for known truths, achieving near-zero latency for verified data.

Innovative Features (V1.6)

  • Deterministic Assembly: Final responses are not generated by the AI directly. Instead, verified Raw Data is physically combined using system-held "Adherents" (language templates). This mechanism aims to achieve 0% hallucination in the final output.

Roadmap

A Micro-MVP is planned for release, demonstrating:

  • Real-time mathematical rejection by a ComplexVerifier.
  • Visibility into RejectionLoop iterations.
  • Proof of SessionPurge efficacy in preventing context drift.

Finetunes
1 models