acyclib 0.3.0

ML library for directed acyclic tensor graphs.
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

use crate::device::{Device, OperationResult};

pub trait CoreDeviceOps: Device {
    fn select(
        batch_size: usize,
        input_batched: bool,
        input_size: usize,
        output_size: usize,
        input: &Self::BufferF32,
        indices: &Self::BufferI32,
        output: &mut Self::BufferF32,
    ) -> OperationResult<Self::DeviceError>;

    fn select_backprop(
        batch_size: usize,
        input_grad_batched: bool,
        input_size: usize,
        output_size: usize,
        indices: &Self::BufferI32,
        output_grad: &Self::BufferF32,
        input_grad: &mut Self::BufferF32,
    ) -> OperationResult<Self::DeviceError>;

    fn softmax_across_batch(
        batch_size: usize,
        single_size: usize,
        input: &Self::BufferF32,
        output: &mut Self::BufferF32,
    ) -> OperationResult<Self::DeviceError>;

    fn crossentropy(
        size: usize,
        pred: &Self::BufferF32,
        target: &Self::BufferF32,
        output: &mut Self::BufferF32,
    ) -> OperationResult<Self::DeviceError>;

    fn backprop_softmax_crossentropy(
        size: usize,
        softmaxed: &Self::BufferF32,
        target: &Self::BufferF32,
        output_grad: &Self::BufferF32,
        input_grad: &mut Self::BufferF32,
    ) -> OperationResult<Self::DeviceError>;
}