ternlang-core 0.3.3

Compiler and VM for Ternlang — balanced ternary language with affirm/tend/reject trit semantics, @sparseskip codegen, and BET bytecode execution.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

// Module:  stdlib/models/mlp.tern
// Purpose: Multi-Layer Perceptron in Ternary
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Standard Feed-Forward network composed entirely of ternary ops.

struct TritMLP {
    layers: int,
    // Simplification for compilation
    w1: trittensor<4 x 4>,
    w2: trittensor<4 x 4>
}

fn mlp_forward(model: TritMLP, input: trittensor<4 x 1>) -> trittensor<4 x 1> {
    @sparseskip
    let h1: trittensor<4 x 1> = model.w1 * input;
    
    @sparseskip
    let out: trittensor<4 x 1> = model.w2 * h1;
    
    return out;
}

fn mlp_train_step(model: TritMLP, input: trittensor<4 x 1>, target: trit) -> trit {
    // Forward, compute loss, backward
    return affirm; // Step completed
}