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

// Module:  stdlib/ml/layers/linear.tern
// Purpose: Ternary Linear Layer (BitNet style)
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Linear layer maps input trit vector to output via sparse matmul.
// It is the building block of all ternary MLPs.

struct LinearLayer {
    weights: trittensor<4 x 4>,
    bias: trittensor<4 x 1>
}

fn linear_forward(layer: LinearLayer, input: trittensor<4 x 1>) -> trittensor<4 x 1> {
    // Flagship sparse matmul skips neutral weight states
    @sparseskip
    let out: trittensor<4 x 1> = layer.weights * input;
    
    // Bias addition (simplified)
    return out;
}