realizar 0.8.4

Pure Rust ML inference engine built from scratch - model serving for GGUF and safetensors
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
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395


// ============================================================================
// EXTREME TDD: Comprehensive Tests
// ============================================================================

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

    // ------------------------------------------------------------------------
    // simd_layer_norm Tests
    // ------------------------------------------------------------------------

    #[test]
    fn test_layer_norm_basic() {
        let input = vec![1.0, 2.0, 3.0, 4.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);

        // Output should have mean ≈ 0
        let mean: f32 = output.iter().sum::<f32>() / output.len() as f32;
        assert!(mean.abs() < 1e-5, "Mean should be ~0, got {}", mean);

        // Output should have std ≈ 1
        let var: f32 = output.iter().map(|x| (x - mean).powi(2)).sum::<f32>() / output.len() as f32;
        let std = var.sqrt();
        assert!((std - 1.0).abs() < 0.01, "Std should be ~1, got {}", std);
    }

    #[test]
    fn test_layer_norm_with_scale() {
        let input = vec![1.0, 2.0, 3.0, 4.0];
        let weight = vec![2.0, 2.0, 2.0, 2.0];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);

        // With scale=2, std should be ~2
        let mean: f32 = output.iter().sum::<f32>() / output.len() as f32;
        let var: f32 = output.iter().map(|x| (x - mean).powi(2)).sum::<f32>() / output.len() as f32;
        let std = var.sqrt();
        assert!((std - 2.0).abs() < 0.01, "Std should be ~2, got {}", std);
    }

    #[test]
    fn test_layer_norm_with_bias() {
        let input = vec![1.0, 2.0, 3.0, 4.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];
        let bias = vec![5.0, 5.0, 5.0, 5.0];
        let output = simd_layer_norm(&input, &weight, Some(&bias), 1e-5);

        // With bias=5, mean should be ~5
        let mean: f32 = output.iter().sum::<f32>() / output.len() as f32;
        assert!((mean - 5.0).abs() < 0.01, "Mean should be ~5, got {}", mean);
    }

    #[test]
    fn test_layer_norm_empty() {
        let input: Vec<f32> = vec![];
        let weight: Vec<f32> = vec![];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);
        assert!(output.is_empty());
    }

    #[test]
    fn test_layer_norm_single_element() {
        let input = vec![5.0];
        let weight = vec![1.0];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);
        // Single element: mean=5, var=0, so normalized = 0
        assert!((output[0]).abs() < 1e-3);
    }

    #[test]
    fn test_layer_norm_uniform_input() {
        let input = vec![3.0, 3.0, 3.0, 3.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);
        // Uniform input: mean=3, var=0+eps, normalized ≈ 0
        for &x in &output {
            assert!(x.abs() < 0.1);
        }
    }

    #[test]
    fn test_layer_norm_negative_values() {
        let input = vec![-2.0, -1.0, 1.0, 2.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);

        // Mean should be 0, values should preserve sign relationship
        assert!(output[0] < output[1]);
        assert!(output[1] < output[2]);
        assert!(output[2] < output[3]);
    }

    #[test]
    fn test_layer_norm_large_values() {
        let input = vec![1000.0, 2000.0, 3000.0, 4000.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];
        let output = simd_layer_norm(&input, &weight, None, 1e-5);

        // Should still have mean ≈ 0 and std ≈ 1
        let mean: f32 = output.iter().sum::<f32>() / output.len() as f32;
        assert!(mean.abs() < 1e-4);
    }

    // ------------------------------------------------------------------------
    // simd_rms_norm Tests
    // ------------------------------------------------------------------------

    #[test]
    fn test_rms_norm_basic() {
        let input = vec![1.0, 2.0, 3.0];
        let weight = vec![1.0, 1.0, 1.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);

        // RMS = sqrt((1 + 4 + 9) / 3) = sqrt(14/3) ≈ 2.16
        let rms = (14.0_f32 / 3.0).sqrt();
        let expected: Vec<f32> = input.iter().map(|x| x / rms).collect();

        for (out, exp) in output.iter().zip(expected.iter()) {
            assert!((out - exp).abs() < 1e-5);
        }
    }

    #[test]
    fn test_rms_norm_with_scale() {
        let input = vec![1.0, 2.0, 3.0];
        let weight = vec![2.0, 2.0, 2.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);

        let rms = (14.0_f32 / 3.0).sqrt();
        let expected: Vec<f32> = input.iter().map(|x| x / rms * 2.0).collect();

        for (out, exp) in output.iter().zip(expected.iter()) {
            assert!((out - exp).abs() < 1e-5);
        }
    }

    #[test]
    fn test_rms_norm_empty() {
        let input: Vec<f32> = vec![];
        let weight: Vec<f32> = vec![];
        let output = simd_rms_norm(&input, &weight, 1e-5);
        assert!(output.is_empty());
    }

    #[test]
    fn test_rms_norm_single_element() {
        let input = vec![5.0];
        let weight = vec![1.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);
        // RMS of [5] = 5, so output = 5/5 = 1
        assert!((output[0] - 1.0).abs() < 1e-5);
    }

    #[test]
    fn test_rms_norm_unit_vector() {
        // For input [1, 0, 0] with weight [1, 1, 1]
        // RMS = sqrt(mean(x^2)) = sqrt(1/3)
        // output = input / RMS * weight = [sqrt(3), 0, 0]
        let input = vec![1.0, 0.0, 0.0];
        let weight = vec![1.0, 1.0, 1.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);

        let expected = 3.0_f32.sqrt(); // sqrt(3) ≈ 1.732
        assert!(
            (output[0] - expected).abs() < 1e-4,
            "Expected {}, got {}",
            expected,
            output[0]
        );
        assert!(output[1].abs() < 1e-5);
        assert!(output[2].abs() < 1e-5);
    }

    #[test]
    fn test_rms_norm_zeros() {
        let input = vec![0.0, 0.0, 0.0];
        let weight = vec![1.0, 1.0, 1.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);

        // RMS = sqrt(eps), output = 0 / sqrt(eps) = 0
        for &x in &output {
            assert!(x.abs() < 1e-2);
        }
    }

    #[test]
    fn test_rms_norm_negative_values() {
        let input = vec![-3.0, 4.0];
        let weight = vec![1.0, 1.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);

        // RMS = sqrt((9 + 16) / 2) = sqrt(12.5) ≈ 3.54
        let rms = (12.5_f32).sqrt();
        assert!((output[0] - (-3.0 / rms)).abs() < 1e-5);
        assert!((output[1] - (4.0 / rms)).abs() < 1e-5);
    }

    #[test]
    fn test_rms_norm_preserves_direction() {
        let input = vec![3.0, 4.0]; // 3-4-5 right triangle
        let weight = vec![1.0, 1.0];
        let output = simd_rms_norm(&input, &weight, 1e-5);

        // Direction should be preserved: output[1]/output[0] = 4/3
        let ratio = output[1] / output[0];
        assert!((ratio - 4.0 / 3.0).abs() < 1e-5);
    }

    // ------------------------------------------------------------------------
    // apply_rope Tests
    // ------------------------------------------------------------------------

    #[test]
    fn test_rope_position_zero() {
        let mut x = vec![1.0, 2.0, 3.0, 4.0]; // 4 hidden, 1 head, head_dim=4
        let original = x.clone();
        apply_rope(&mut x, 4, 1, 0, 10000.0);

        // At position 0, angle = 0, cos(0) = 1, sin(0) = 0
        // So output should equal input
        for (out, orig) in x.iter().zip(original.iter()) {
            assert!((out - orig).abs() < 1e-5);
        }
    }

    #[test]
    fn test_rope_rotation_property() {
        let mut x = vec![1.0, 0.0, 0.0, 1.0]; // 4 hidden, 1 head
        apply_rope(&mut x, 4, 1, 1, 10000.0);

        // After rotation, magnitude should be preserved for each pair
        let mag0 = (x[0] * x[0] + x[2] * x[2]).sqrt();
        let mag1 = (x[1] * x[1] + x[3] * x[3]).sqrt();

        assert!((mag0 - 1.0).abs() < 1e-5, "Magnitude of pair 0 should be 1");
        assert!((mag1 - 1.0).abs() < 1e-5, "Magnitude of pair 1 should be 1");
    }

    #[test]
    fn test_rope_multiple_heads() {
        let mut x = vec![1.0; 8]; // 8 hidden, 2 heads, head_dim = 4
        let original = x.clone();
        apply_rope(&mut x, 8, 2, 0, 10000.0);

        // At position 0, should be unchanged
        for (out, orig) in x.iter().zip(original.iter()) {
            assert!((out - orig).abs() < 1e-5);
        }
    }

    #[test]
    fn test_rope_different_positions() {
        let mut x1 = vec![1.0; 4];
        let mut x2 = vec![1.0; 4];

        apply_rope(&mut x1, 4, 1, 0, 10000.0);
        apply_rope(&mut x2, 4, 1, 1, 10000.0);

        // Different positions should give different results
        assert!((x1[0] - x2[0]).abs() > 1e-6 || (x1[1] - x2[1]).abs() > 1e-6);
    }

    #[test]
    fn test_rope_theta_scaling() {
        let mut x1 = vec![1.0; 4];
        let mut x2 = vec![1.0; 4];

        apply_rope(&mut x1, 4, 1, 10, 10000.0);
        apply_rope(&mut x2, 4, 1, 10, 1000.0);

        // Different theta affects higher frequency dimensions (i > 0)
        // For i=0, freq = 1/theta^0 = 1 (same regardless of theta)
        // For i=1, freq = 1/theta^(2/head_dim) which differs by theta
        // So check x[1] or x[3] (the second pair uses i=1)
        assert!(
            (x1[1] - x2[1]).abs() > 1e-5 || (x1[3] - x2[3]).abs() > 1e-5,
            "Different theta should give different results for non-zero frequency indices"
        );
    }

    #[test]
    fn test_rope_large_position() {
        let mut x = vec![1.0, 2.0, 3.0, 4.0];
        apply_rope(&mut x, 4, 1, 1000, 10000.0);

        // Results should be finite
        for &val in &x {
            assert!(val.is_finite());
        }
    }

    #[test]
    fn test_rope_eight_heads() {
        let hidden_dim = 64;
        let num_heads = 8;
        let mut x = vec![0.5; hidden_dim];

        apply_rope(&mut x, hidden_dim, num_heads, 5, 10000.0);

        // All values should be finite
        for &val in &x {
            assert!(val.is_finite());
        }
    }

    #[test]
    fn test_rope_preserves_length() {
        let mut x = vec![3.0, 4.0, 0.0, 0.0]; // pairs: (3,0), (4,0)
        apply_rope(&mut x, 4, 1, 1, 10000.0);

        assert_eq!(x.len(), 4);
    }

    // ------------------------------------------------------------------------
    // Integration Tests
    // ------------------------------------------------------------------------

    #[test]
    fn test_norm_then_rope() {
        let input = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0];
        let weight = vec![1.0; 8];

        // First normalize
        let normalized = simd_rms_norm(&input, &weight, 1e-5);

        // Then apply RoPE
        let mut output = normalized;
        apply_rope(&mut output, 8, 2, 5, 10000.0);

        // Results should be finite
        for &val in &output {
            assert!(val.is_finite());
        }
    }

    #[test]
    fn test_layer_norm_vs_rms_norm() {
        let input = vec![1.0, 2.0, 3.0, 4.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];

        let ln_output = simd_layer_norm(&input, &weight, None, 1e-5);
        let rms_output = simd_rms_norm(&input, &weight, 1e-5);

        // LayerNorm centers (mean=0), RMSNorm doesn't
        let ln_mean: f32 = ln_output.iter().sum::<f32>() / 4.0;
        let rms_mean: f32 = rms_output.iter().sum::<f32>() / 4.0;

        assert!(ln_mean.abs() < 1e-5, "LayerNorm should have mean ~0");
        assert!(rms_mean.abs() > 0.1, "RMSNorm should not center");
    }

    // ------------------------------------------------------------------------
    // Edge Cases
    // ------------------------------------------------------------------------

    #[test]
    fn test_layer_norm_eps_impact() {
        let input = vec![0.0, 0.0, 0.0, 0.0];
        let weight = vec![1.0, 1.0, 1.0, 1.0];

        // With var=0, eps prevents division by zero
        let output = simd_layer_norm(&input, &weight, None, 1e-5);
        for &val in &output {
            assert!(val.is_finite());
        }
    }

    #[test]
    fn test_rms_norm_eps_impact() {
        let input = vec![0.0, 0.0];
        let weight = vec![1.0, 1.0];

        // With sum_sq=0, eps prevents division by zero
        let output = simd_rms_norm(&input, &weight, 1e-5);
        for &val in &output {
            assert!(val.is_finite());
        }
    }

    #[test]
    fn test_rope_half_dim_calculation() {
        // Test with various head dimensions
        for (hidden_dim, num_heads) in [(8, 2), (16, 4), (32, 8), (64, 16)] {
            let mut x = vec![1.0; hidden_dim];
            apply_rope(&mut x, hidden_dim, num_heads, 1, 10000.0);

            // Should not panic and should produce finite values
            for &val in &x {
                assert!(val.is_finite());
            }
        }
    }
}