dfdx 0.13.0

Ergonomic auto differentiation in Rust, with pytorch like apis.
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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227

use num_traits::Float;
use rand_distr::{uniform::SampleUniform, Uniform};

use crate::{shapes::*, tensor::*, tensor_ops::*};

use super::*;

pub mod builder {
    #[derive(Debug)]
    pub struct Embedding<const VOCAB: usize, const DIM: usize>;
}

impl<const V: usize, const M: usize, E: Dtype, D: Device<E>> BuildOnDevice<D, E>
    for builder::Embedding<V, M>
where
    Embedding<V, M, E, D>: BuildModule<D, E>,
{
    type Built = Embedding<V, M, E, D>;
    fn try_build_on_device(device: &D) -> Result<Self::Built, D::Err> {
        Self::Built::try_build(device)
    }
}

/// An embedding
/// Initializes [Self::weight] from a Uniform distribution
/// between [-1 / sqrt(I), 1 / sqrt(I)].
///
/// # Generics
/// - `VOCAB` The size of the vocabulary, inputs integer values must be between
///    0 and VOCAB;
/// - `DIM` The "output" size of vectors & matrices which are the vectors being selected.
///
/// # Examples
/// `Embedding<5, 2>` can act on vectors with SEQ integer elements (with values between 0 and 4), and results in a SEQ tensor of
/// usually f32 elements being the rows in [Self::weight].
/// ```rust
///
/// # use dfdx::prelude::*;
/// # let dev: Cpu = Default::default();
/// type Model = Embedding<7, 2>;
/// let mut model = dev.build_module::<Model, f32>();
/// // single sequence of ids
/// let inputs: Tensor<Rank1<5>, usize, _> = dev.zeros();
/// let _: Tensor<(Const<5>, Const<2>,), f32, _> = model.forward(inputs);
/// // Dynamic sequence of ids
/// let inputs: Tensor<(usize, ), usize, _> = dev.zeros_like(&(5, ));
/// let _: Tensor<(usize, Const<2>,), f32, _> = model.forward(inputs);
/// // batched sequence of ids
/// let inputs: Tensor<Rank2<10, 5>, usize, _> = dev.zeros();
/// let _: Tensor<(Const<10>, Const<5>, Const<2>), f32, _> = model.forward(inputs);
/// ```
#[derive(Debug, Clone)]
pub struct Embedding<const VOCAB: usize, const DIM: usize, E: Dtype, D: Storage<E>> {
    /// Transposed weight matrix, shape (I, O)
    pub weight: Tensor<Rank2<VOCAB, DIM>, E, D>,
}

impl<const V: usize, const M: usize, E: Dtype, D: Storage<E>> NonMutableModule
    for Embedding<V, M, E, D>
{
}

impl<const C: usize, const M: usize, E: Dtype + Float + SampleUniform, D: Device<E>>
    TensorCollection<E, D> for Embedding<C, M, E, D>
{
    type To<E2: Dtype, D2: Device<E2>> = Embedding<C, M, E2, D2>;

    fn iter_tensors<V: ModuleVisitor<Self, E, D>>(
        visitor: &mut V,
    ) -> Result<Option<Self::To<V::E2, V::D2>>, V::Err> {
        visitor.visit_fields(
            Self::tensor(
                "weight",
                |s| &s.weight,
                |s| &mut s.weight,
                TensorOptions::reset_with(|t| {
                    let b: E = E::ONE / E::from_usize(C).unwrap().sqrt();
                    t.try_fill_with_distr(Uniform::new(-b, b))
                }),
            ),
            |weight| Embedding { weight },
        )
    }
}

impl<const V: usize, const M: usize, SEQ: Dim, E: Dtype, D: Device<E>, T: Tape<E, D>>
    Module<Tensor<(SEQ,), usize, D, T>> for Embedding<V, M, E, D>
{
    type Output = Tensor<(SEQ, Const<M>), E, D, T>;
    type Error = D::Err;

    fn try_forward(&self, input: Tensor<(SEQ,), usize, D, T>) -> Result<Self::Output, D::Err> {
        let (input, tape) = input.split_tape();
        self.weight.clone().put_tape(tape).try_gather(input)
    }
}

impl<
        const VOCAB: usize,
        const DIM: usize,
        BATCH: Dim,
        SEQ: Dim,
        E: Dtype,
        D: Device<E>,
        T: Tape<E, D>,
    > Module<Tensor<(BATCH, SEQ), usize, D, T>> for Embedding<VOCAB, DIM, E, D>
{
    type Output = Tensor<(BATCH, SEQ, Const<DIM>), E, D, T>;
    type Error = D::Err;

    fn try_forward(
        &self,
        input: Tensor<(BATCH, SEQ), usize, D, T>,
    ) -> Result<Self::Output, D::Err> {
        let (input, tape) = input.split_tape();
        self.weight.clone().put_tape(tape).try_gather(input)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::tests::*;

    const W: [[f64; 5]; 2] = [
        [-0.3458893, -0.30371523, -0.3712057, 0.14303583, -0.0268966],
        [0.11733949, 0.14059687, -0.10670426, -0.09373143, 0.18974298],
    ];

    #[test]
    fn test_embedding_initialize() {
        let dev: TestDevice = Default::default();
        let m = dev.build_module::<builder::Embedding<2000, 1>, TestDtype>();
        let bound: TestDtype = NumCast::from(1.0 / (2000.0.sqrt())).unwrap();
        for v in m.weight.as_vec() {
            assert!(-bound <= v && v <= bound && v != TestDtype::zero());
        }
    }

    #[test]
    fn embedding_forward_1d() {
        let dev: TestDevice = Default::default();

        let model = Embedding {
            weight: dev.tensor(W),
        }
        .to_dtype::<TestDtype>();

        let x = dev.tensor([0, 0, 1]);
        let y = model.forward(x.leaky_trace());
        assert_close_to_literal!(
            y,
            [
                [-0.3458893, -0.30371523, -0.3712057, 0.14303583, -0.0268966],
                [-0.3458893, -0.30371523, -0.3712057, 0.14303583, -0.0268966],
                [0.11733949, 0.14059687, -0.10670426, -0.09373143, 0.18974298],
            ]
        );

        let g = y.square().mean().backward();
        assert_close_to_literal!(
            g.get(&model.weight),
            [
                [
                    -0.09223715,
                    -0.08099073,
                    -0.09898819,
                    0.03814289,
                    -0.007172427,
                ],
                [
                    0.015645266,
                    0.01874625,
                    -0.014227235,
                    -0.012497525,
                    0.025299065,
                ],
            ]
        );
    }

    #[test]
    fn test_forward_2d() {
        let dev: TestDevice = Default::default();

        let model = Embedding {
            weight: dev.tensor(W),
        };

        let x = dev.tensor([[0, 0], [0, 1]]);
        let y = model.forward(x.leaky_trace());
        assert_close_to_literal!(
            y,
            [
                [
                    [-0.3458893, -0.30371523, -0.3712057, 0.14303583, -0.0268966],
                    [-0.3458893, -0.30371523, -0.3712057, 0.14303583, -0.0268966],
                ],
                [
                    [-0.3458893, -0.30371523, -0.3712057, 0.14303583, -0.0268966],
                    [0.11733949, 0.14059687, -0.10670426, -0.09373143, 0.18974298],
                ],
            ]
        );

        let g = y.square().mean().backward();
        assert_close_to_literal!(
            g.get(&model.weight),
            [
                [
                    -0.103766784,
                    -0.091114566,
                    -0.11136171,
                    0.042910747,
                    -0.008068981,
                ],
                [
                    0.011733949,
                    0.014059687,
                    -0.010670426,
                    -0.009373143,
                    0.018974299,
                ],
            ]
        );
    }
}