Crate ternsig 

Ternsig - TernarySignal Foundation

The foundational crate that unlocked a new way of thinking about neural learning. TernarySignals (polarity + magnitude) replace floating-point weights entirely.

Core Components

• TernarySignal: The fundamental unit (polarity ∈ {-1,0,+1}, magnitude ∈ 0-255)
• TensorISA: Hot-reloadable neural network instruction set (.tisa.asm files)
• Adaptive Learning: Mastery learning - 23ms updates, 90% accuracy threshold
• Thermogram Integration: Persistent weight storage with temperature lifecycle

Three-Tier Learning System

1. Tier 1 (Priors): Offline instinct creation via mastery learning
2. Tier 2 (Coordination): SNN + neuromodulators gate learning
3. Tier 3 (Runtime): Continuous 23ms adaptive refinement

Design Principles

• No floats: All weights are TernarySignal (2 bytes each)
• CPU-only: Integer arithmetic, no GPU required
• Persistent: All weights use thermograms (survive crashes)
• Hot-reloadable: .tisa.asm files define network architecture

Example

use ternsig::{TernarySignal, TensorInterpreter, assemble};

// Load chip definition
let program = assemble(include_str!("onset.tisa.asm"))?;
let mut interpreter = TensorInterpreter::new(&program)?;

// Forward pass with ternary weights
interpreter.set_input(&input_signals);
interpreter.execute()?;
let output = interpreter.get_output();

Re-exports

pub use tensor_isa::TensorInterpreter;

pub use tensor_isa::TensorInstruction;

pub use tensor_isa::TensorAction;

pub use tensor_isa::TensorRegister;

pub use tensor_isa::HotBuffer;

pub use tensor_isa::ColdBuffer;

pub use tensor_isa::TensorDtype;

pub use tensor_isa::TensorModifier;

pub use tensor_isa::assemble;

pub use tensor_isa::AssembledProgram;

pub use tensor_isa::AssemblerError;

pub use tensor_isa::serialize_tisa;

pub use tensor_isa::deserialize_tisa;

pub use tensor_isa::load_tisa_file;

pub use tensor_isa::save_tisa_file;

pub use tensor_isa::HotReloadManager;

pub use tensor_isa::ReloadableInterpreter;

pub use tensor_isa::ArchStats;

pub use tensor_isa::ModEvent;

pub use tensor_isa::ShapeSpec;

pub use tensor_isa::WireSpec;

pub use tensor_isa::WireType;

pub use learning::PolarityState;

pub use learning::SurpriseOptimizer;

pub use learning::SurpriseOptimizerConfig;

pub use learning::FloatingTernaryLayer;

pub use learning::OptimizerStats;

pub use thermo::TensorThermogram;

pub use thermo::WeightContent;

pub use thermo::ThermalState;

pub use thermo::TensorContent;

pub use thermo::ProgramMetaContent;

pub use thermo::LearningStateContent;

Modules

learning

Adaptive Learning System for TernarySignal Weights

tensor_isa

Tensor ISA - Instruction Set Architecture for Neural Network Definitions

thermo

Thermogram Bridge for TensorISA

Structs

TernarySignal

Ternary signal: polarity + magnitude

Enums

TernsigError

Ternsig error type