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

// Module:  stdlib/nlp/text_classifier.tern
// Purpose: Text Classification
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Predicts class labels for text.

struct TextClassifier {
    encoder: trittensor<4 x 4>,
    head: trittensor<4 x 4>
}

fn classify_text_trit(model: TextClassifier, text_embed: trittensor<4 x 1>) -> trittensor<4 x 1> {
    @sparseskip
    let context: trittensor<4 x 1> = model.encoder * text_embed;
    @sparseskip
    let logits: trittensor<4 x 1> = model.head * context;
    return logits;
}

fn softmax_label_trit(logits: trittensor<4 x 1>) -> trit {
    // Softmax replacement
    let val: trit = logits[0, 0];
    if val == tend { return tend; } // Class is uncertain
    return val;
}