Skip to content

tensorrent/tent-io

Repository files navigation

TENT — Tensor Engine for Nondeterministic Transcription

Bradley Wallace — Independent Researcher Licensed under the Sovereign Integrity Protocol (SIP) v1.1


Overview

TENT is a deterministic tensor engine that replaces TensorFlow/PyTorch's floating-point pipeline with exact integer arithmetic. Built on the Big Reveal Architecture (BRA), TENT routes tensors through a 32³ voxel lattice using gravitational field equations, boustrophedon siphon traversal, and the Wallace Transform.

TENT eliminates:

  • Floating-point drift → all operations are exact integer
  • Stochastic gradient descent → replaced by deterministic constraint satisfaction (RC Stack)
  • Black-box inference → every decision has a traceable integer charge path

Architecture

Core Engine

File Size Purpose
src/tent_v9.py 68K Core TENT v9 engine — production tensor routing
src/tent_v9_production.py 70K Deployment-hardened production variant
src/tent_v10_vixel.py 28K Vixel-integrated engine with scroll memory
src/tent_v10_pipeline.py 18K Streaming pipeline for continuous inference
src/bra_bridge.py 5K Python↔Rust FFI bridge
src/bra_kernel.rs 4K Rust BRA kernel (compiled to cdylib)

BRA (Big Reveal Architecture)

EigenCharge triplets (hash, trace, det) for every data structure:

hash  → FNV-1a (u64)      → Identity fingerprinting
trace → Linear F369        → Within-class compactness
det   → Quadratic F369     → Between-class separation

Formally equivalent to MCR² (Maximal Coding Rate Reduction) from representation learning, operating entirely in integer space.

Wallace Transform

  1. F369 Table — Pre-computed integer lookup (12,000 entries)
  2. Ulam Spiral Addressing — Storage addresses on prime-number spiral
  3. Boustrophedon Siphon — Alternating-direction traversal (every voxel visited exactly once)

Key Properties

  • ADC Resolution: 10¹⁴ bins (vs float's ~10⁷) — eliminates charge collisions
  • Resonance Scoring: 3-tier exact match (0/1/2) replaces continuous similarity
  • Deterministic Reproducibility: Same input → same output, always, on any hardware
  • RC Stack Integration: Every inference passes through constraint gates RC1–RC14

Directory Structure

tent-io/
├── src/
│   ├── tent_v9.py                # Core engine (68K)
│   ├── tent_v9_production.py     # Production variant (70K)
│   ├── tent_v10_vixel.py         # Vixel integration (28K)
│   ├── tent_v10_pipeline.py      # Streaming pipeline (18K)
│   ├── bra_bridge.py             # Python↔Rust FFI (5K)
│   └── bra_kernel.rs             # Rust BRA kernel (4K)
├── tests/
│   └── tent_tests.py             # Test suite
├── papers/
│   ├── tent_v91_formal.pdf       # Formal specification
│   ├── tent_v91_spec.pdf         # Technical specification
│   ├── trinity_core_BRA.pdf      # BRA architecture paper
│   └── trinity_core_BRA.tex      # LaTeX source
├── build.sh                      # Rust kernel build script
└── SIP_LICENSE.md

Quick Start

chmod +x build.sh && ./build.sh
python3 tests/tent_tests.py
python3 -c "from src.tent_v9 import TENTEngine; print('TENT loaded')"

Related Repositories


License

Copyright (c) 2026, Bradley Wallace (tensorrent). All rights reserved.

SIP License v1.1 — Personal/educational: royalty-free. Commercial: prohibited without prior license. Unlicensed commercial use triggers 8.4% perpetual gross profit penalty.

See SIP_LICENSE.md for full terms.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors