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


// ========================================================================
// IMP-114: True GPU Batched GEMM Kernel Tests
// ========================================================================

#[test]
#[cfg(feature = "gpu")]
#[serial_test::serial]
fn test_imp_114a_flattened_batched_gemm_correctness() {
    // IMP-114a: Verify flattened batched GEMM computes correct results
    // Strategy: Flatten [batch, m, k] @ [batch, k, n] into single large matmul
    let config = GGUFConfig {
        architecture: "test".to_string(),
        constraints: crate::gguf::ArchConstraints::from_architecture("test"),
        hidden_dim: 64,
        intermediate_dim: 128,
        num_layers: 1,
        num_heads: 4,
        num_kv_heads: 4,
        vocab_size: 50,
        context_length: 128,
        rope_theta: 10000.0,
        eps: 1e-5,
        rope_type: 0,
            explicit_head_dim: None,
            query_pre_attn_scalar: None,
        bos_token_id: None,
            eos_token_id: None,
    };

    let model = create_test_model_with_config(&config);
    let cached_model = OwnedQuantizedModelCached::new(model);

    let batch_size = 4;
    let m = 8;
    let k = 16;
    let n = 8;

    // Create batched matrices
    let batched_a: Vec<f32> = (0..batch_size * m * k)
        .map(|i| ((i % 13) as f32 - 6.0) * 0.1)
        .collect();
    let batched_b: Vec<f32> = (0..batch_size * k * n)
        .map(|i| ((i % 11) as f32 - 5.0) * 0.1)
        .collect();

    // Use flattened batched GEMM (true single dispatch)
    let result = cached_model
        .flattened_batched_gemm(&batched_a, &batched_b, batch_size, m, k, n)
        .expect("Flattened batched GEMM should succeed");

    // Output should be [batch_size, m, n]
    assert_eq!(
        result.len(),
        batch_size * m * n,
        "IMP-114a: Output should have shape [batch, m, n]"
    );

    // Verify by computing reference per-batch
    for b in 0..batch_size {
        let a_start = b * m * k;
        let b_start = b * k * n;
        let out_start = b * m * n;

        for i in 0..m {
            for j in 0..n {
                let mut expected = 0.0f32;
                for kk in 0..k {
                    expected += batched_a[a_start + i * k + kk] * batched_b[b_start + kk * n + j];
                }
                let actual = result[out_start + i * n + j];
                let diff = (expected - actual).abs();
                assert!(
                    diff < 1e-3,
                    "IMP-114a: Batch {} mismatch at ({},{}): expected={}, actual={}, diff={}",
                    b,
                    i,
                    j,
                    expected,
                    actual,
                    diff
                );
            }
        }
    }
}

#[test]
#[cfg(feature = "gpu")]
#[serial_test::serial]
fn test_imp_114b_flattened_matches_loop() {
    // IMP-114b: Verify flattened approach matches loop-based approach
    let config = GGUFConfig {
        architecture: "test".to_string(),
        constraints: crate::gguf::ArchConstraints::from_architecture("test"),
        hidden_dim: 64,
        intermediate_dim: 128,
        num_layers: 1,
        num_heads: 8,
        num_kv_heads: 8,
        vocab_size: 50,
        context_length: 128,
        rope_theta: 10000.0,
        eps: 1e-5,
        rope_type: 0,
            explicit_head_dim: None,
            query_pre_attn_scalar: None,
        bos_token_id: None,
            eos_token_id: None,
    };

    let model = create_test_model_with_config(&config);
    let cached_model = OwnedQuantizedModelCached::new(model);

    let batch_size = 8;
    let m = 16;
    let k = 8;
    let n = 16;

    let batched_a: Vec<f32> = (0..batch_size * m * k)
        .map(|i| ((i % 17) as f32 - 8.0) * 0.05)
        .collect();
    let batched_b: Vec<f32> = (0..batch_size * k * n)
        .map(|i| ((i % 19) as f32 - 9.0) * 0.05)
        .collect();

    // Loop-based (IMP-113)
    let loop_result = cached_model
        .batched_gemm_single_dispatch(&batched_a, &batched_b, batch_size, m, k, n)
        .expect("Loop GEMM should succeed");

    // Flattened (IMP-114)
    let flat_result = cached_model
        .flattened_batched_gemm(&batched_a, &batched_b, batch_size, m, k, n)
        .expect("Flattened GEMM should succeed");

    assert_eq!(loop_result.len(), flat_result.len());
    for i in 0..loop_result.len() {
        let diff = (loop_result[i] - flat_result[i]).abs();
        assert!(
            diff < 1e-3,
            "IMP-114b: Results differ at {}: loop={}, flat={}, diff={}",
            i,
            loop_result[i],
            flat_result[i],
            diff
        );
    }
}

#[test]
#[cfg(feature = "gpu")]
#[serial_test::serial]
fn test_imp_114c_flattened_attention_correctness() {
    // IMP-114c: Verify flattened attention matches reference
    let config = GGUFConfig {
        architecture: "test".to_string(),
        constraints: crate::gguf::ArchConstraints::from_architecture("test"),
        hidden_dim: 64,
        intermediate_dim: 128,
        num_layers: 1,
        num_heads: 4,
        num_kv_heads: 4,
        vocab_size: 50,
        context_length: 128,
        rope_theta: 10000.0,
        eps: 1e-5,
        rope_type: 0,
            explicit_head_dim: None,
            query_pre_attn_scalar: None,
        bos_token_id: None,
            eos_token_id: None,
    };

    let model = create_test_model_with_config(&config);
    let cached_model = OwnedQuantizedModelCached::new(model);

    let seq_len = 8;
    let hidden_dim = config.hidden_dim;

    let q: Vec<f32> = (0..seq_len * hidden_dim)
        .map(|i| ((i % 13) as f32 - 6.0) * 0.1)
        .collect();
    let k: Vec<f32> = (0..seq_len * hidden_dim)
        .map(|i| ((i % 11) as f32 - 5.0) * 0.1)
        .collect();
    let v: Vec<f32> = (0..seq_len * hidden_dim)
        .map(|i| ((i % 7) as f32 - 3.0) * 0.1)
        .collect();

    // Reference (IMP-113 single dispatch)
    let reference = cached_model
        .single_dispatch_multihead_attention(&q, &k, &v, seq_len)
        .expect("Reference attention should succeed");

    // Flattened (IMP-114)
    let result = cached_model
        .flattened_multihead_attention(&q, &k, &v, seq_len)
        .expect("Flattened attention should succeed");

    assert_eq!(result.len(), reference.len());
    for i in 0..result.len() {
        let diff = (result[i] - reference[i]).abs();
        assert!(
            diff < 1e-3,
            "IMP-114c: Attention differs at {}: ref={}, flat={}, diff={}",
            i,
            reference[i],
            result[i],
            diff
        );
    }
}

#[test]
#[cfg(feature = "gpu")]
#[serial_test::serial]
fn test_imp_114d_large_batch_flattened() {
    // IMP-114d: Test with larger batch sizes where flattening benefits
    let config = GGUFConfig {
        architecture: "test".to_string(),
        constraints: crate::gguf::ArchConstraints::from_architecture("test"),
        hidden_dim: 128,
        intermediate_dim: 256,
        num_layers: 1,
        num_heads: 16, // Larger number of heads
        num_kv_heads: 16,
        vocab_size: 50,
        context_length: 256,
        rope_theta: 10000.0,
        eps: 1e-5,
        rope_type: 0,
            explicit_head_dim: None,
            query_pre_attn_scalar: None,
        bos_token_id: None,
            eos_token_id: None,
    };

    let model = create_test_model_with_config(&config);
    let cached_model = OwnedQuantizedModelCached::new(model);

    let batch_size = 16;
    let m = 8;
    let k = 8;
    let n = 8;

    let batched_a: Vec<f32> = (0..batch_size * m * k)
        .map(|i| ((i % 23) as f32 - 11.0) * 0.04)
        .collect();
    let batched_b: Vec<f32> = (0..batch_size * k * n)
        .map(|i| ((i % 29) as f32 - 14.0) * 0.04)
        .collect();

    let result = cached_model
        .flattened_batched_gemm(&batched_a, &batched_b, batch_size, m, k, n)
        .expect("Large batch flattened GEMM should succeed");

    assert_eq!(
        result.len(),
        batch_size * m * n,
        "IMP-114d: Output should have correct dimensions"
    );

    // Verify non-trivial output
    let sum: f32 = result.iter().map(|x| x.abs()).sum();
    assert!(
        sum > 0.01,
        "IMP-114d: Output should have non-trivial values, got sum={}",
        sum
    );
}