aprender-serve 0.64.0

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


#[test]
fn test_gelu_monotonicity_positive_range() {
    // GELU should be monotonically increasing for x > 0
    let values: Vec<f32> = (0..10).map(|i| i as f32 * 0.5).collect();
    let input = Tensor::from_vec(vec![10], values).expect("input");
    let output = gelu(&input).expect("gelu");

    for i in 1..10 {
        assert!(
            output.data()[i] >= output.data()[i - 1],
            "GELU should be monotonic for positive x"
        );
    }
}

#[test]
fn test_gelu_approximation_accuracy() {
    // Test that GELU approximation is reasonably accurate
    let test_points = [0.0, 0.5, 1.0, 2.0, -0.5, -1.0];

    for &x in &test_points {
        let input = Tensor::from_vec(vec![1], vec![x]).expect("input");
        let output = gelu(&input).expect("gelu");

        // GELU(0) = 0
        if x == 0.0 {
            assert!(output.data()[0].abs() < 1e-6);
        }
        // GELU(x) > 0 for x > 0
        if x > 0.0 {
            assert!(output.data()[0] > 0.0);
        }
        // GELU should be finite
        assert!(output.data()[0].is_finite());
    }
}

// =============================================================================
// Error Path Coverage: Construction Errors
// =============================================================================

#[test]
fn test_linear_zero_in_features_error() {
    let result = Linear::new(0, 8);
    assert!(result.is_err());

    if let Err(RealizarError::InvalidShape { reason }) = result {
        assert!(reason.contains("> 0") || reason.contains("in_features"));
    }
}

#[test]
fn test_linear_zero_out_features_error() {
    let result = Linear::new(8, 0);
    assert!(result.is_err());

    if let Err(RealizarError::InvalidShape { reason }) = result {
        assert!(reason.contains("> 0") || reason.contains("out_features"));
    }
}

#[test]
fn test_layer_norm_zero_normalized_shape_error() {
    let result = LayerNorm::new(0, 1e-5);
    assert!(result.is_err());

    if let Err(RealizarError::InvalidShape { reason }) = result {
        assert!(reason.contains("> 0") || reason.contains("normalized_shape"));
    }
}

#[test]
fn test_fused_layer_norm_linear_zero_feature_dim_error() {
    let result = FusedLayerNormLinear::new(0, 8, 1e-5);
    assert!(result.is_err());

    if let Err(RealizarError::InvalidShape { reason }) = result {
        assert!(reason.contains("> 0") || reason.contains("feature_dim"));
    }
}

#[test]
fn test_fused_layer_norm_linear_zero_out_features_error() {
    let result = FusedLayerNormLinear::new(8, 0, 1e-5);
    assert!(result.is_err());

    if let Err(RealizarError::InvalidShape { reason }) = result {
        assert!(reason.contains("> 0") || reason.contains("out_features"));
    }
}

#[test]
fn test_ffn_zero_hidden_dim_error() {
    let result = FeedForward::new(0, 16);
    assert!(result.is_err());
}

#[test]
fn test_ffn_zero_intermediate_dim_error() {
    let result = FeedForward::new(4, 0);
    assert!(result.is_err());
}

// =============================================================================
// Batched Input Edge Cases
// =============================================================================

#[test]
fn test_linear_single_batch_dimension() {
    let linear = Linear::new(4, 8).expect("create");

    // Single item batch: [1, 4]
    let input = Tensor::from_vec(vec![1, 4], vec![1.0, 2.0, 3.0, 4.0]).expect("input");
    let output = linear.forward(&input).expect("forward");

    assert_eq!(output.shape(), &[1, 8]);
}

#[test]
fn test_layer_norm_single_batch() {
    let layer_norm = LayerNorm::new(4, 1e-5).expect("create");

    // Single item batch: [1, 4]
    let input = Tensor::from_vec(vec![1, 4], vec![1.0, 2.0, 3.0, 4.0]).expect("input");
    let output = layer_norm.forward(&input).expect("forward");

    assert_eq!(output.shape(), &[1, 4]);
}

#[test]
fn test_fused_layer_norm_linear_single_batch() {
    let fused = FusedLayerNormLinear::new(4, 8, 1e-5).expect("create");

    // Single item batch
    let input = Tensor::from_vec(vec![1, 4], vec![1.0, 2.0, 3.0, 4.0]).expect("input");

    let serial = fused.forward(&input).expect("serial");
    let parallel = fused.forward_parallel(&input).expect("parallel");

    assert_eq!(serial.shape(), &[1, 8]);
    assert_eq!(parallel.shape(), &[1, 8]);
}

// =============================================================================
// Property-Based Style Tests
// =============================================================================

#[test]
fn test_softmax_output_bounds() {
    // Softmax output should be in [0, 1] for all inputs
    let test_values: Vec<f32> = (-10..10).map(|i| i as f32 * 0.5).collect();
    let input = Tensor::from_vec(vec![20], test_values).expect("input");
    let output = softmax(&input).expect("softmax");

    for &val in output.data() {
        assert!(val >= 0.0, "Softmax output should be >= 0");
        assert!(val <= 1.0, "Softmax output should be <= 1");
    }
}

#[test]
fn test_layer_norm_output_mean_zero() {
    // LayerNorm with default gamma=1, beta=0 should produce mean ~= 0
    let layer_norm = LayerNorm::new(8, 1e-5).expect("create");

    // Various input distributions
    let test_inputs = [
        vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0],
        vec![-4.0, -3.0, -2.0, -1.0, 1.0, 2.0, 3.0, 4.0],
        vec![100.0, 100.1, 100.2, 100.3, 100.4, 100.5, 100.6, 100.7],
    ];

    for values in test_inputs {
        let input = Tensor::from_vec(vec![8], values).expect("input");
        let output = layer_norm.forward(&input).expect("forward");

        let mean: f32 = output.data().iter().sum::<f32>() / 8.0;
        assert!(
            mean.abs() < 1e-4,
            "LayerNorm output mean should be ~0, got {}",
            mean
        );
    }
}

#[test]
fn test_linear_zero_input_produces_bias() {
    let mut linear = Linear::new(4, 3).expect("create");

    // Set all weights to 0, set specific bias
    for w in linear.weight_mut().iter_mut() {
        *w = 0.0;
    }
    linear.bias_mut()[0] = 1.0;
    linear.bias_mut()[1] = 2.0;
    linear.bias_mut()[2] = 3.0;

    // Zero input should produce bias
    let input = Tensor::from_vec(vec![4], vec![0.0, 0.0, 0.0, 0.0]).expect("input");
    let output = linear.forward(&input).expect("forward");

    assert!((output.data()[0] - 1.0).abs() < 1e-6);
    assert!((output.data()[1] - 2.0).abs() < 1e-6);
    assert!((output.data()[2] - 3.0).abs() < 1e-6);
}

// =============================================================================
// RoPE and Position Embedding Coverage
// =============================================================================

#[test]
fn test_rope_creation_various_dimensions() {
    for dim in [4, 8, 16, 32, 64, 128] {
        let rope = RoPE::new(dim, 512).expect("create");
        assert_eq!(rope.head_dim(), dim);
        assert_eq!(rope.max_seq_len(), 512);
    }
}

#[test]
fn test_scaled_rope_scaling_types() {
    // Test all scaling types can be created
    let linear = ScaledRoPE::new(64, 2048, RopeScalingType::Linear { scale: 2.0 }).expect("linear");
    assert_eq!(linear.head_dim(), 64);

    let dynamic =
        ScaledRoPE::new(64, 2048, RopeScalingType::Dynamic { scale: 2.0 }).expect("dynamic");
    assert_eq!(dynamic.head_dim(), 64);

    let ntk = ScaledRoPE::new(
        64,
        2048,
        RopeScalingType::NTKAware {
            scale: 2.0,
            alpha: 1.0,
        },
    )
    .expect("ntk");
    assert_eq!(ntk.head_dim(), 64);

    let yarn = ScaledRoPE::new(
        64,
        2048,
        RopeScalingType::YaRN {
            scale: 2.0,
            original_max_seq_len: 4096,
            attention_factor: 1.0,
            beta_fast: 32.0,
            beta_slow: 1.0,
        },
    )
    .expect("yarn");
    assert_eq!(yarn.head_dim(), 64);
}

#[test]
fn test_alibi_creation_various_heads() {
    for num_heads in [1, 2, 4, 8, 16, 32] {
        let alibi = ALiBi::new(num_heads).expect("create");
        assert_eq!(alibi.num_heads(), num_heads);
    }
}

#[test]
fn test_rope_forward_shape_preservation() {
    let rope = RoPE::new(8, 512).expect("create");

    // Test various input shapes
    let shapes = [vec![4, 8], vec![2, 4, 8], vec![1, 1, 8]];

    for shape in shapes {
        let size: usize = shape.iter().product();
        let input = Tensor::from_vec(shape.clone(), vec![0.1f32; size]).expect("input");
        let output = rope.forward(&input, 0).expect("forward");

        assert_eq!(output.shape(), &shape[..], "RoPE should preserve shape");
    }
}

#[test]
fn test_alibi_get_bias_various_seq_lengths() {
    let alibi = ALiBi::new(4).expect("create");

    for seq_len in [1, 4, 16, 64, 128] {
        let bias = alibi.get_bias(seq_len).expect("get_bias");

        // Bias shape should be [num_heads, seq_len, seq_len]
        assert_eq!(bias.shape(), &[4, seq_len, seq_len]);
    }
}

// =============================================================================
// Attention Submodule Additional Coverage
// =============================================================================

#[test]
fn test_attention_zero_head_dim_error() {
    let result = Attention::new(0);
    assert!(result.is_err(), "Should error on zero head_dim");
}

#[test]
fn test_attention_large_head_dim() {
    let attn = Attention::new(256).expect("create");

    let q = Tensor::from_vec(vec![2, 256], vec![0.1f32; 512]).expect("q");
    let k = Tensor::from_vec(vec![2, 256], vec![0.1f32; 512]).expect("k");
    let v = Tensor::from_vec(vec![2, 256], vec![0.1f32; 512]).expect("v");

    let output = attn.forward(&q, &k, &v).expect("forward");
    assert_eq!(output.shape(), &[2, 256]);
}

#[test]
fn test_sliding_window_attention_window_larger_than_seq() {
    let swa = SlidingWindowAttention::new(8, 100).expect("create");

    // seq_len=4 < window_size=100
    let q = Tensor::from_vec(vec![4, 8], vec![0.1f32; 32]).expect("q");
    let k = Tensor::from_vec(vec![4, 8], vec![0.1f32; 32]).expect("k");
    let v = Tensor::from_vec(vec![4, 8], vec![0.1f32; 32]).expect("v");

    // Should work - effectively full attention
    let output = swa.forward(&q, &k, &v).expect("forward");
    assert_eq!(output.shape(), &[4, 8]);
}

#[test]
fn test_multi_head_attention_weight_accessors() {
    let mut mha = MultiHeadAttention::mha(64, 8).expect("create");

    // Test all weight accessors are accessible and writable
    let w_q = mha.w_q_mut();
    assert!(!w_q.is_empty());
    w_q[0] = 1.0;

    let w_k = mha.w_k_mut();
    assert!(!w_k.is_empty());
    w_k[0] = 2.0;

    let w_v = mha.w_v_mut();
    assert!(!w_v.is_empty());
    w_v[0] = 3.0;

    let w_o = mha.w_o_mut();
    assert!(!w_o.is_empty());
    w_o[0] = 4.0;
}