briny_ai 0.5.0

A tiny & efficient AI inference engine
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
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170

use super::exp;
use libm::erff;

use crate::nn::{
    tensors::{Tensor, WithGrad},
    TensorFloat,
};
use tensor_optim::TensorOps;

#[cfg(feature = "dyntensor")]
use alloc::vec;

#[cfg(feature = "alloc")]
use alloc::boxed::Box;
#[cfg(not(feature = "alloc"))]
use box_closure::{Align8, OpaqueFn};

/// Performs GELU activation.
#[must_use]
#[cfg(feature = "dyntensor")]
pub fn gelu(
    input: &WithGrad<Tensor<TensorFloat>>,
) -> (
    Tensor<TensorFloat>,
    Box<dyn Fn(Tensor<TensorFloat>) -> Tensor<TensorFloat> + '_>,
) {
    let shape = input.get_value().shape();
    let input_data = input.get_value().data();

    let mut out_data = vec![0.0; input_data.len()];
    let mut phi_data = vec![0.0; input_data.len()];

    let inv_sqrt2 = 0.707_106_781_186_547_6; // 1 / sqrt(2)

    out_data
        .iter_mut()
        .zip(phi_data.iter_mut())
        .zip(input_data.iter())
        .for_each(|((y, phi), &x)| {
            let cdf = 0.5 * TensorFloat::from(1.0 + erff((x * inv_sqrt2) as f32));
            *phi = cdf;
            *y = x * cdf;
        });

    let out = Tensor::new(shape, &out_data);

    let back = move |grad_output: Tensor<TensorFloat>| {
        let dy = grad_output.data();
        let mut grad = vec![TensorFloat::default(); dy.len()];

        let inv_sqrt2pi = TensorFloat::from(0.398_942_280_401_432_7); // 1 / sqrt(2π)

        grad.iter_mut()
            .zip(input_data.iter())
            .zip(phi_data.iter())
            .zip(dy.iter())
            .for_each(|(((g, &x), &phi), &dyi)| {
                let pdf = inv_sqrt2pi * exp(-TensorFloat::from(0.5) * x * x);
                *g = dyi * (phi + x * pdf);
            });

        Tensor::new(shape, &grad)
    };

    (out, Box::new(back))
}

/// Performs GELU activation.
#[must_use]
#[cfg(all(feature = "alloc", not(feature = "dyntensor")))]
pub fn gelu<const N: usize, const D: usize>(
    input: &WithGrad<Tensor<TensorFloat, N, D>>,
) -> (
    Tensor<TensorFloat, N, D>,
    Box<dyn Fn(Tensor<TensorFloat, N, D>) -> Tensor<TensorFloat, N, D> + '_>,
) {
    use tensor_optim::ConstTensorOps;

    let shape: &[usize; D] = input.get_value().shape_array();
    let input_data = input.get_value().data();

    let mut out_data = [TensorFloat::default(); N];
    let mut phi_data = [TensorFloat::default(); N];

    let inv_sqrt2 = TensorFloat::from(0.707_106_781_186_547_6);

    out_data
        .iter_mut()
        .zip(phi_data.iter_mut())
        .zip(input_data.iter())
        .for_each(|((y, phi), &x)| {
            let cdf = 0.5 * TensorFloat::from(1.0 + erff((x * inv_sqrt2) as f32));
            *phi = cdf;
            *y = x * cdf;
        });

    let out = Tensor::new(shape, &out_data);

    let back = move |grad_output: Tensor<TensorFloat, N, D>| {
        let dy = grad_output.data();
        let mut grad = [TensorFloat::default(); N];

        let inv_sqrt2pi = TensorFloat::from(0.398_942_280_401_432_7);

        grad.iter_mut()
            .zip(input_data.iter())
            .zip(phi_data.iter())
            .zip(dy.iter())
            .for_each(|(((g, &x), &phi), &dyi)| {
                let pdf = inv_sqrt2pi * exp(-TensorFloat::from(0.5) * x * x);
                *g = dyi * (phi + x * pdf);
            });

        Tensor::new(shape, &grad)
    };

    (out, Box::new(back))
}

/// Performs GELU activation.
#[must_use]
#[cfg(not(feature = "alloc"))]
pub fn gelu<const N: usize, const D: usize>(
    input: &WithGrad<Tensor<TensorFloat, N, D>>,
) -> (
    Tensor<TensorFloat, N, D>,
    OpaqueFn<'_, Tensor<TensorFloat, N, D>, Tensor<TensorFloat, N, D>, Align8<128>>,
) {
    use tensor_optim::ConstTensorOps;

    let shape: &[usize; D] = input.get_value().shape_array();
    let input_data = input.get_value().data();

    let mut out_data = [0.0; N];
    let mut phi_data = [0.0; N];

    let inv_sqrt2 = 0.707_106_781_186_547_6;

    out_data
        .iter_mut()
        .zip(phi_data.iter_mut())
        .zip(input_data.iter())
        .for_each(|((y, phi), &x)| {
            let cdf = 0.5 * TensorFloat::from(1.0 + erff((x * inv_sqrt2) as f32));
            *phi = cdf;
            *y = x * cdf;
        });

    let out = Tensor::new(shape, &out_data);

    let back = move |grad_output: Tensor<TensorFloat, N, D>| {
        let dy = grad_output.data();
        let mut grad = [0.0; N];

        let inv_sqrt2pi = 0.398_942_280_401_432_7;

        grad.iter_mut()
            .zip(input_data.iter())
            .zip(phi_data.iter())
            .zip(dy.iter())
            .for_each(|(((g, &x), &phi), &dyi)| {
                let pdf = inv_sqrt2pi * exp(-0.5 * x * x);
                *g = dyi * (phi + x * pdf);
            });

        Tensor::new(shape, &grad)
    };

    (out, OpaqueFn::new(back))
}