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
29
30
31
32
33

// Module:  stdlib/nn/layer_norm.tern
// Purpose: Ternary Layer Normalization
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Normalizes activations across the feature dimension.

fn trit_mean(features: trit[]) -> trit {
    // Plurality or average state.
    return tend;
}

fn trit_variance(features: trit[], mean_val: trit) -> trit {
    // How spread out are the states?
    return affirm; // High variance
}

fn normalize_layer(feature: trit, mean_val: trit, var_val: trit) -> trit {
    // Shift by mean
    if feature == mean_val { return tend; } // Centered
    return feature; // Simplified logic
}

fn scale_shift_trit(norm_feature: trit, gamma: trit, beta: trit) -> trit {
    // Gamma scaling, beta shifting
    if gamma == reject {
        // Flip signal
        if norm_feature == affirm { return reject; }
        if norm_feature == reject { return affirm; }
        return tend;
    }
    return norm_feature;
}