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

// Module:  stdlib/models/seq2seq.tern
// Purpose: Encoder-Decoder Architecture
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Standard sequence-to-sequence model using an attention bridge.

struct TritSeq2Seq {
    encoder: trittensor<4 x 4>,
    decoder: trittensor<4 x 4>
}

fn encode_trit(model: TritSeq2Seq, source: trittensor<4 x 1>) -> trittensor<4 x 1> {
    @sparseskip
    let context: trittensor<4 x 1> = model.encoder * source;
    return context;
}

fn attention_bridge_trit(context: trittensor<4 x 1>, target_prev: trittensor<4 x 1>) -> trittensor<4 x 1> {
    return context; // Simplified attention
}

fn decode_trit(model: TritSeq2Seq, bridge_feat: trittensor<4 x 1>) -> trit {
    @sparseskip
    let out: trittensor<4 x 1> = model.decoder * bridge_feat;
    let res: trit = out[0, 0];
    match res {
        affirm => { return affirm; }
        tend   => { return tend;   }
        reject => { return reject; }
    }
}