trustformers-core 0.1.1

Core traits and utilities for TrustformeRS
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
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252

/// Tests for activation functions: gelu, relu, sigmoid, tanh, silu, swiglu
#[cfg(test)]
mod tests {
    use crate::ops::activations::{gelu, gelu_new, relu, sigmoid, silu, swiglu, tanh};
    use crate::tensor::Tensor;
    use scirs2_core::ndarray::ArrayD;

    fn make_f32_tensor(data: Vec<f32>, shape: Vec<usize>) -> Tensor {
        let arr = ArrayD::from_shape_vec(shape, data).expect("Valid shape");
        Tensor::F32(arr)
    }

    fn extract_f32(t: &Tensor) -> Vec<f32> {
        match t {
            Tensor::F32(arr) => arr.iter().cloned().collect(),
            _ => panic!("Expected F32 tensor"),
        }
    }

    // ---- ReLU tests ----

    #[test]
    fn test_relu_positive_values_unchanged() {
        let t = make_f32_tensor(vec![1.0, 2.0, 3.0], vec![3]);
        let result = relu(&t).expect("relu should succeed");
        let data = extract_f32(&result);
        assert!((data[0] - 1.0).abs() < 1e-6);
        assert!((data[1] - 2.0).abs() < 1e-6);
        assert!((data[2] - 3.0).abs() < 1e-6);
    }

    #[test]
    fn test_relu_negative_values_become_zero() {
        let t = make_f32_tensor(vec![-1.0, -5.0, -0.001], vec![3]);
        let result = relu(&t).expect("relu should succeed");
        let data = extract_f32(&result);
        for v in data {
            assert_eq!(v, 0.0);
        }
    }

    #[test]
    fn test_relu_zero_stays_zero() {
        let t = make_f32_tensor(vec![0.0], vec![1]);
        let result = relu(&t).expect("relu should succeed");
        let data = extract_f32(&result);
        assert_eq!(data[0], 0.0);
    }

    #[test]
    fn test_relu_mixed_values() {
        let t = make_f32_tensor(vec![-2.0, 0.0, 3.0, -0.5, 1.0], vec![5]);
        let result = relu(&t).expect("relu should succeed");
        let data = extract_f32(&result);
        assert_eq!(data[0], 0.0);
        assert_eq!(data[1], 0.0);
        assert!((data[2] - 3.0).abs() < 1e-6);
        assert_eq!(data[3], 0.0);
        assert!((data[4] - 1.0).abs() < 1e-6);
    }

    #[test]
    fn test_relu_2d_tensor() {
        let t = make_f32_tensor(vec![-1.0, 2.0, -3.0, 4.0], vec![2, 2]);
        let result = relu(&t).expect("relu should succeed");
        let data = extract_f32(&result);
        assert_eq!(data[0], 0.0);
        assert!((data[1] - 2.0).abs() < 1e-6);
        assert_eq!(data[2], 0.0);
        assert!((data[3] - 4.0).abs() < 1e-6);
    }

    // ---- Sigmoid tests ----

    #[test]
    fn test_sigmoid_zero_gives_half() {
        let t = make_f32_tensor(vec![0.0], vec![1]);
        let result = sigmoid(&t).expect("sigmoid should succeed");
        let data = extract_f32(&result);
        assert!((data[0] - 0.5).abs() < 1e-6);
    }

    #[test]
    fn test_sigmoid_large_positive_near_one() {
        let t = make_f32_tensor(vec![20.0], vec![1]);
        let result = sigmoid(&t).expect("sigmoid should succeed");
        let data = extract_f32(&result);
        assert!(data[0] > 0.99);
        assert!(data[0] <= 1.0);
    }

    #[test]
    fn test_sigmoid_large_negative_near_zero() {
        let t = make_f32_tensor(vec![-20.0], vec![1]);
        let result = sigmoid(&t).expect("sigmoid should succeed");
        let data = extract_f32(&result);
        assert!(data[0] < 0.01);
        assert!(data[0] >= 0.0);
    }

    #[test]
    fn test_sigmoid_output_range_all_in_0_1() {
        let mut s = 42u64;
        let mut vals = Vec::new();
        for _ in 0..20 {
            s = s.wrapping_mul(6364136223846793005).wrapping_add(1442695040888963407);
            let v = (s % 1000) as f32 / 100.0 - 5.0; // [-5, 5)
            vals.push(v);
        }
        let t = make_f32_tensor(vals.clone(), vec![vals.len()]);
        let result = sigmoid(&t).expect("sigmoid should succeed");
        let data = extract_f32(&result);
        for v in data {
            assert!(
                (0.0..=1.0).contains(&v),
                "Sigmoid output {} not in [0, 1]",
                v
            );
        }
    }

    // ---- Tanh tests ----

    #[test]
    fn test_tanh_zero_gives_zero() {
        let t = make_f32_tensor(vec![0.0], vec![1]);
        let result = tanh(&t).expect("tanh should succeed");
        let data = extract_f32(&result);
        assert!((data[0]).abs() < 1e-6);
    }

    #[test]
    fn test_tanh_range_minus1_to_1() {
        let vals: Vec<f32> = vec![-10.0, -1.0, 0.0, 1.0, 10.0];
        let n = vals.len();
        let t = make_f32_tensor(vals, vec![n]);
        let result = tanh(&t).expect("tanh should succeed");
        let data = extract_f32(&result);
        for v in data {
            assert!((-1.0..=1.0).contains(&v));
        }
    }

    #[test]
    fn test_tanh_odd_function() {
        let t_pos = make_f32_tensor(vec![1.5], vec![1]);
        let t_neg = make_f32_tensor(vec![-1.5], vec![1]);
        let r_pos = tanh(&t_pos).expect("tanh should succeed");
        let r_neg = tanh(&t_neg).expect("tanh should succeed");
        let pos_data = extract_f32(&r_pos);
        let neg_data = extract_f32(&r_neg);
        assert!((pos_data[0] + neg_data[0]).abs() < 1e-5);
    }

    // ---- SiLU (Swish) tests ----

    #[test]
    fn test_silu_zero_gives_zero() {
        let t = make_f32_tensor(vec![0.0], vec![1]);
        let result = silu(&t).expect("silu should succeed");
        let data = extract_f32(&result);
        assert!((data[0]).abs() < 1e-6);
    }

    #[test]
    fn test_silu_large_positive_approaches_identity() {
        // silu(x) = x * sigmoid(x) ≈ x for large x
        let t = make_f32_tensor(vec![10.0], vec![1]);
        let result = silu(&t).expect("silu should succeed");
        let data = extract_f32(&result);
        assert!((data[0] - 10.0).abs() < 0.1);
    }

    #[test]
    fn test_silu_negative_input_small_output() {
        // silu(-10) ≈ -10 * sigmoid(-10) ≈ 0
        let t = make_f32_tensor(vec![-10.0], vec![1]);
        let result = silu(&t).expect("silu should succeed");
        let data = extract_f32(&result);
        assert!(data[0].abs() < 0.01);
    }

    // ---- GELU tests ----

    #[test]
    fn test_gelu_large_positive_approaches_identity() {
        let t = make_f32_tensor(vec![10.0], vec![1]);
        let result = gelu(&t).expect("gelu should succeed");
        let data = extract_f32(&result);
        // gelu(10) ≈ 10
        assert!((data[0] - 10.0).abs() < 0.1);
    }

    #[test]
    fn test_gelu_large_negative_approaches_zero() {
        let t = make_f32_tensor(vec![-10.0], vec![1]);
        let result = gelu(&t).expect("gelu should succeed");
        let data = extract_f32(&result);
        // gelu(-10) ≈ 0
        assert!(data[0].abs() < 0.01);
    }

    #[test]
    fn test_gelu_no_nan() {
        let vals: Vec<f32> = vec![-5.0, -2.0, 0.0, 0.5, 2.0, 5.0];
        let n = vals.len();
        let t = make_f32_tensor(vals, vec![n]);
        let result = gelu(&t).expect("gelu should succeed");
        let data = extract_f32(&result);
        for v in data {
            assert!(!v.is_nan(), "gelu output was NaN");
            assert!(!v.is_infinite(), "gelu output was infinite");
        }
    }

    #[test]
    fn test_gelu_new_no_nan() {
        let vals: Vec<f32> = vec![-3.0, -1.0, 0.0, 1.0, 3.0];
        let n = vals.len();
        let t = make_f32_tensor(vals, vec![n]);
        let result = gelu_new(&t).expect("gelu_new should succeed");
        let data = extract_f32(&result);
        for v in data {
            assert!(!v.is_nan());
        }
    }

    // ---- SwiGLU tests ----

    #[test]
    fn test_swiglu_basic() {
        let x = make_f32_tensor(vec![1.0, 2.0, 3.0], vec![3]);
        let gate = make_f32_tensor(vec![1.0, 1.0, 1.0], vec![3]);
        // SwiGLU(x, gate) = SiLU(gate) * x
        let result = swiglu(&x, &gate).expect("swiglu should succeed");
        let data = extract_f32(&result);
        // silu(1.0) * 1.0 = 1/(1+e^-1) * 1.0 ≈ 0.731
        assert!((data[0] - 0.731).abs() < 0.01);
    }

    #[test]
    fn test_swiglu_zero_gate_gives_zero() {
        let x = make_f32_tensor(vec![5.0, 10.0], vec![2]);
        let gate = make_f32_tensor(vec![-100.0, -100.0], vec![2]);
        let result = swiglu(&x, &gate).expect("swiglu should succeed");
        let data = extract_f32(&result);
        // silu(-100) ≈ 0, so SwiGLU ≈ 0
        for v in data {
            assert!(v.abs() < 0.01);
        }
    }
}