candle-core 0.10.2

Minimalist ML framework.
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

use crate::{Result, Tensor};

#[macro_export]
macro_rules! test_device {
    // TODO: Switch to generating the two last arguments automatically once concat_idents is
    // stable. https://github.com/rust-lang/rust/issues/29599
    ($fn_name: ident, $test_cpu: ident, $test_cuda: ident, $test_metal: ident) => {
        #[test]
        fn $test_cpu() -> Result<()> {
            $fn_name(&Device::Cpu)
        }

        #[cfg(feature = "cuda")]
        #[test]
        fn $test_cuda() -> Result<()> {
            $fn_name(&Device::new_cuda(0)?)
        }

        #[cfg(feature = "metal")]
        #[test]
        fn $test_metal() -> Result<()> {
            $fn_name(&Device::new_metal(0)?)
        }
    };
}

pub fn assert_tensor_eq(t1: &Tensor, t2: &Tensor) -> Result<()> {
    assert_eq!(t1.shape(), t2.shape());
    // Default U8 may not be large enough to hold the sum (`t.sum_all` defaults to the dtype of `t`)
    let eq_tensor = t1.eq(t2)?.to_dtype(crate::DType::U32)?;
    let all_equal = eq_tensor.sum_all()?;
    assert_eq!(all_equal.to_scalar::<u32>()?, eq_tensor.elem_count() as u32);
    Ok(())
}

pub fn to_vec0_round(t: &Tensor, digits: i32) -> Result<f32> {
    let b = 10f32.powi(digits);
    let t = t.to_vec0::<f32>()?;
    Ok(f32::round(t * b) / b)
}

pub fn to_vec1_round(t: &Tensor, digits: i32) -> Result<Vec<f32>> {
    let b = 10f32.powi(digits);
    let t = t.to_vec1::<f32>()?;
    let t = t.iter().map(|t| f32::round(t * b) / b).collect();
    Ok(t)
}

pub fn to_vec2_round(t: &Tensor, digits: i32) -> Result<Vec<Vec<f32>>> {
    let b = 10f32.powi(digits);
    let t = t.to_vec2::<f32>()?;
    let t = t
        .iter()
        .map(|t| t.iter().map(|t| f32::round(t * b) / b).collect())
        .collect();
    Ok(t)
}

pub fn to_vec3_round(t: &Tensor, digits: i32) -> Result<Vec<Vec<Vec<f32>>>> {
    let b = 10f32.powi(digits);
    let t = t.to_vec3::<f32>()?;
    let t = t
        .iter()
        .map(|t| {
            t.iter()
                .map(|t| t.iter().map(|t| f32::round(t * b) / b).collect())
                .collect()
        })
        .collect();
    Ok(t)
}