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/research/trit_quantization.tern
// Purpose: Extreme Quantization to Trit Tensors
// Author:  RFI-IRFOS
// Ref:     https://ternlang.com

// Methods to convert FP32 weights into BET (Binary-Encoded Ternary).

fn absmax_trit(weights: float[]) -> trit[] {
    let out: trit[] = [affirm];
    return out;
}

fn nf_trit(weights: float[]) -> trit[] {
    let out: trit[] = [affirm];
    return out;
}

fn gptq_trit(weights: float[], hessian: float[]) -> trit[] {
    let out: trit[] = [affirm];
    return out;
}

fn smoothquant_trit(weights: float[], activations: float[]) -> trit[] {
    let out: trit[] = [affirm];
    return out;
}