Skip to content

About

Non-stochastic constraint engine enforcing zero-drift linguistic containment, Pāṇinian kāraka logit masking, and deterministic inference guarantees across LLMs and autoregressive token streams.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

TIP Logo

immortal-inference:
Non-Stochastic AI Determinism & Logit Containment Engine

Developed by Joshua St. Germain | TheImmortalPython Labs


License: AGPLv3 License: Commercial CI Status CLA Enforced Python Version

Dual-licensed under AGPLv3 (open-source) and Commercial Terms (proprietary/SaaS). See Section 5.


1. Overview

immortal-inference is a high-performance, non-stochastic constraint engine engineered to enforce zero-drift linguistic containment and deterministic execution guarantees over generative autoregressive models.

By intercepting Hugging Face LogitsProcessor pipelines and PyTorch matrix operations in real time, the engine applies dynamic binary distribution masks prior to final token sampling.


[ Context Input ] ---> [ Model Forward Pass ] ---> [ Pāṇinian Kāraka Mask ] ---> [ Guaranteed Output ]

2. Core Architecture

  • Pāṇinian Kāraka Logit Masking: Maps linguistic dependency structures into mathematical matrices to enforce structural compliance at token step $t+1$.
  • Tensor-Topology Constraint Calculus: Evaluates the token vocabulary space across a constrained topological subspace, setting non-compliant logits to $-\infty$.
  • Deterministic Execution: Guarantees absolute structural compliance without output drift, hallucination, or syntax violations.

3. Performance & Benchmarks

The immortal-inference engine enforces strict token-level determinism using Pāṇinian Kāraka dependency lattices. Long-horizon evaluations over a 10,000-token test horizon demonstrate zero syntactic drift or containment violations.

Metric Base Model Masked Engine
Syntactic Accuracy 50.0% 100.0%
Structural Hallucinations 50.0% 0.0%
Avg Inference Latency 0.00ms 0.67ms
Deterministic Compliance 0.0% 100.0%

Raw telemetry data supporting these metrics is available in PKLM_10k_Horizon_Benchmark_Data.csv.

4. Directory Structure


immortal-inference/
├── .github/workflows/     # CI automation & CLA assistant automation
├── docs/                  # Documents
├── examples/              # Quickstart notebooks & implementation examples
├── pklm/                  # Core logit processor & PyTorch masking implementation
├── signatures/            # Contributor License Agreement (CLA) signature registry
├── tests/                 # Unit & integration verification test suites
├── .gitignore             # Git ignore rules
├── CONTRIBUTING.md        # Contributor guidelines and CLA workflow
├── LICENSE                # GNU Affero General Public License v3.0 (AGPLv3)
├── LICENSE-COMMERCIAL.md  # Commercial License Exception & AGPLv3 Waiver
├── PKLM_10k_Horizon_B...  # 10k-token long-horizon benchmark telemetry dataset
├── PRICING-TIERS.md       # Enterprise Commercial Pricing & Licensing Tiers
├── README.md              # Project overview, architecture, & Section 13 notice
├── SECURITY.md            # Security policy & vulnerability reporting guidelines
├── SOVEREIGN_EXCEPTION.md # Public Sector & Critical Infrastructure Safe Harbor
├── TIP_Logo_Square.jpg    # Enterprise brand logo asset
└── setup.py               # Standard setuptools installation & metadata script

5. Licensing, Author Intent, Community Contributions & Commercial Licensing

This software is dual-licensed, released under the GNU Affero General Public License v3.0 (AGPLv3) for open-source use, with Commercial Licenses available for proprietary deployment.

Standard Open-Source Licensing

Under the AGPLv3, you are free to inspect, modify, and redistribute immortal-inference provided that all modifications and derivative works remain open-source under the same terms.

Explicit Author Intent Notice (Section 13)

Pursuant to Section 13 of the AGPLv3 ("Remote Network Interaction"), the author Joshua St. Germain (TheImmortalPython Labs) explicitly clarifies that any remote interaction with this software—including API access, web service routing, cloud execution, or integration within an LLM inference pipeline—constitutes interaction over a network.

Any entity hosting, wrapping, or exposing the capabilities of immortal-inference through an API or network protocol MUST make the complete source code of the modified or integrated application available to all users receiving such service.

Community Contributions & CLA

All community contributions to immortal-inference require a signed Contributor License Agreement (CLA). Upon submitting a Pull Request, you will be prompted automatically by our workflow to authorize the agreement. See CONTRIBUTING.md for full guidelines.

Commercial Licensing

To utilize immortal-inference in proprietary, closed-source, or SaaS environments without AGPLv3 copyleft obligations, you must acquire a commercial license. Refer to LICENSE-COMMERCIAL.md and PRICING-TIERS.md for terms.


Note: All software assets, source code repositories, and commercial licenses are offered strictly on an as-is basis as self-service intellectual property acquisitions. Transactions convey rights to the technology asset and codebase only; they do not include ongoing consulting services, custom engineering implementations, software-level support desks, or managed engineering deployments.


Monetizing immortal-inference (and Non-Stochastic AI)

To explore how to monetize immortal-inference, read The Monetization Playbook.


Founder & Principal AI Architect: Joshua St. Germain
Enterprise Governance & IP Incubator: TheImmortalPython Labs (theimmortalpythonlabs@proton.me)
Repositories: immortal-inference | the-nsai-guide


© TheImmortalPython Labs. All rights reserved.

About

Non-stochastic constraint engine enforcing zero-drift linguistic containment, Pāṇinian kāraka logit masking, and deterministic inference guarantees across LLMs and autoregressive token streams.

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages