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
350
351
352


#[test]
fn test_phase33_loader_get_tensor_unsupported_qtype() {
    // Use a fake quantization type that doesn't exist
    let mut data = build_gguf_header(1, 0);
    data.extend(build_tensor_info("test", &[32], 255, 0)); // 255 = invalid qtype

    // Pad to 32-byte alignment
    let current_len = data.len();
    let aligned = current_len.div_ceil(32) * 32;
    data.resize(aligned, 0);
    data.extend([0u8; 64]); // dummy data

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let result = model.get_tensor_f32("test", &data);
    assert!(result.is_err());
    let err = result.unwrap_err().to_string();
    assert!(
        err.contains("255") || err.contains("Unsupported"),
        "Error: {}",
        err
    );
}

#[test]
fn test_phase33_loader_get_tensor_q2_k() {
    use crate::gguf::GGUF_TYPE_Q2_K;

    // Q2_K: 84 bytes per 256 elements (super-block)
    let mut data = build_gguf_header(1, 0);
    data.extend(build_tensor_info("test", &[256], GGUF_TYPE_Q2_K, 0));

    // Pad to 32-byte alignment
    let current_len = data.len();
    let aligned = current_len.div_ceil(32) * 32;
    data.resize(aligned, 0);

    // Add Q2_K super-block (84 bytes)
    data.extend([0u8; 84]);

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let result = model.get_tensor_f32("test", &data);
    assert!(
        result.is_ok(),
        "Q2_K extraction should work: {:?}",
        result.err()
    );
    let tensor = result.expect("test value should be present");
    assert_eq!(tensor.len(), 256);
}

#[test]
fn test_phase33_loader_get_tensor_q4_k() {
    use crate::gguf::GGUF_TYPE_Q4_K;

    // Q4_K: 144 bytes per 256 elements (super-block)
    let mut data = build_gguf_header(1, 0);
    data.extend(build_tensor_info("test", &[256], GGUF_TYPE_Q4_K, 0));

    // Pad to 32-byte alignment
    let current_len = data.len();
    let aligned = current_len.div_ceil(32) * 32;
    data.resize(aligned, 0);

    // Add Q4_K super-block (144 bytes)
    data.extend([0u8; 144]);

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let result = model.get_tensor_f32("test", &data);
    assert!(
        result.is_ok(),
        "Q4_K extraction should work: {:?}",
        result.err()
    );
    let tensor = result.expect("test value should be present");
    assert_eq!(tensor.len(), 256);
}

#[test]
fn test_phase33_loader_get_tensor_q5_k() {
    use crate::gguf::GGUF_TYPE_Q5_K;

    // Q5_K: 176 bytes per 256 elements (super-block)
    let mut data = build_gguf_header(1, 0);
    data.extend(build_tensor_info("test", &[256], GGUF_TYPE_Q5_K, 0));

    // Pad to 32-byte alignment
    let current_len = data.len();
    let aligned = current_len.div_ceil(32) * 32;
    data.resize(aligned, 0);

    // Add Q5_K super-block (176 bytes)
    data.extend([0u8; 176]);

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let result = model.get_tensor_f32("test", &data);
    assert!(
        result.is_ok(),
        "Q5_K extraction should work: {:?}",
        result.err()
    );
    let tensor = result.expect("test value should be present");
    assert_eq!(tensor.len(), 256);
}

#[test]
fn test_phase33_loader_get_tensor_q6_k() {
    use crate::gguf::GGUF_TYPE_Q6_K;

    // Q6_K: 210 bytes per 256 elements (super-block)
    let mut data = build_gguf_header(1, 0);
    data.extend(build_tensor_info("test", &[256], GGUF_TYPE_Q6_K, 0));

    // Pad to 32-byte alignment
    let current_len = data.len();
    let aligned = current_len.div_ceil(32) * 32;
    data.resize(aligned, 0);

    // Add Q6_K super-block (210 bytes)
    data.extend([0u8; 210]);

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let result = model.get_tensor_f32("test", &data);
    assert!(
        result.is_ok(),
        "Q6_K extraction should work: {:?}",
        result.err()
    );
    let tensor = result.expect("test value should be present");
    assert_eq!(tensor.len(), 256);
}

// =============================================================================
// RoPE Type Tests
// =============================================================================

#[test]
fn test_phase33_loader_rope_type_llama() {
    let mut data = build_gguf_header(0, 1);
    let arch_value = build_gguf_string("llama");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch_value));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.rope_type(), Some(0)); // NORM style for LLaMA
}

#[test]
fn test_phase33_loader_rope_type_qwen2() {
    let mut data = build_gguf_header(0, 1);
    let arch_value = build_gguf_string("qwen2");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch_value));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.rope_type(), Some(2)); // NEOX style for Qwen2
}

#[test]
fn test_phase33_loader_rope_type_phi3() {
    let mut data = build_gguf_header(0, 1);
    let arch_value = build_gguf_string("phi3");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch_value));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.rope_type(), Some(2)); // NEOX style for Phi3
}

#[test]
fn test_phase33_loader_rope_type_from_scaling() {
    let mut data = build_gguf_header(0, 2);
    let arch_value = build_gguf_string("custom");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch_value));
    let yarn_value = build_gguf_string("yarn");
    data.extend(build_gguf_metadata(
        "custom.rope.scaling.type",
        8,
        &yarn_value,
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.rope_type(), Some(2)); // NEOX from yarn scaling
}

#[test]
fn test_phase33_loader_rope_type_linear_scaling() {
    let mut data = build_gguf_header(0, 2);
    let arch_value = build_gguf_string("custom");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch_value));
    let linear_value = build_gguf_string("linear");
    data.extend(build_gguf_metadata(
        "custom.rope.scaling.type",
        8,
        &linear_value,
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.rope_type(), Some(0)); // NORM from linear scaling
}

// =============================================================================
// Token ID Tests
// =============================================================================

#[test]
fn test_phase33_loader_bos_token_id() {
    let mut data = build_gguf_header(0, 1);
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.bos_token_id",
        4,
        &1u32.to_le_bytes(),
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.bos_token_id(), Some(1));
}

#[test]
fn test_phase33_loader_bos_token_id_missing() {
    let data = build_gguf_header(0, 0);
    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.bos_token_id(), None);
}

#[test]
fn test_phase33_loader_eos_token_id() {
    let mut data = build_gguf_header(0, 1);
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.eos_token_id",
        4,
        &2u32.to_le_bytes(),
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.eos_token_id(), Some(2));
}

// =============================================================================
// Vocabulary Tests
// =============================================================================

#[test]
fn test_phase33_loader_vocabulary() {
    let mut data = build_gguf_header(0, 1);
    // Array of strings: element_type (8=string) + array_len (3) + strings
    let mut array_bytes = Vec::new();
    array_bytes.extend_from_slice(&8u32.to_le_bytes()); // element type: string
    array_bytes.extend_from_slice(&3u64.to_le_bytes()); // array length: 3
    array_bytes.extend(build_gguf_string("hello"));
    array_bytes.extend(build_gguf_string("world"));
    array_bytes.extend(build_gguf_string("!"));
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.tokens",
        9,
        &array_bytes,
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let vocab = model.vocabulary().expect("Should have vocabulary");
    assert_eq!(vocab.len(), 3);
    assert_eq!(vocab[0], "hello");
    assert_eq!(vocab[1], "world");
    assert_eq!(vocab[2], "!");
}

#[test]
fn test_phase33_loader_vocabulary_missing() {
    let data = build_gguf_header(0, 0);
    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert!(model.vocabulary().is_none());
}

// =============================================================================
// Decode Tests
// =============================================================================

#[test]
fn test_phase33_loader_decode_basic() {
    let mut data = build_gguf_header(0, 1);
    let mut array_bytes = Vec::new();
    array_bytes.extend_from_slice(&8u32.to_le_bytes());
    array_bytes.extend_from_slice(&3u64.to_le_bytes());
    array_bytes.extend(build_gguf_string("hello"));
    array_bytes.extend(build_gguf_string("▁world")); // SentencePiece space
    array_bytes.extend(build_gguf_string("!"));
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.tokens",
        9,
        &array_bytes,
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let decoded = model.decode(&[0, 1, 2]);
    assert_eq!(decoded, "hello world!");
}

#[test]
fn test_phase33_loader_decode_no_vocab() {
    let data = build_gguf_header(0, 0);
    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let decoded = model.decode(&[65, 66, 67]); // ASCII fallback
    assert_eq!(decoded, "ABC");
}

#[test]
fn test_phase33_loader_decode_byte_tokens() {
    let mut data = build_gguf_header(0, 1);
    let mut array_bytes = Vec::new();
    array_bytes.extend_from_slice(&8u32.to_le_bytes());
    array_bytes.extend_from_slice(&2u64.to_le_bytes());
    array_bytes.extend(build_gguf_string("<0x48>")); // 'H'
    array_bytes.extend(build_gguf_string("<0x69>")); // 'i'
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.tokens",
        9,
        &array_bytes,
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let decoded = model.decode(&[0, 1]);
    assert_eq!(decoded, "Hi");
}

// =============================================================================
// Encode Tests
// =============================================================================

#[test]
fn test_phase33_loader_encode_basic() {
    let mut data = build_gguf_header(0, 1);
    let mut array_bytes = Vec::new();
    array_bytes.extend_from_slice(&8u32.to_le_bytes());
    array_bytes.extend_from_slice(&3u64.to_le_bytes());
    array_bytes.extend(build_gguf_string("▁hello")); // SentencePiece prefix
    array_bytes.extend(build_gguf_string("▁world"));
    array_bytes.extend(build_gguf_string("!"));
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.tokens",
        9,
        &array_bytes,
    ));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let tokens = model.encode("hello world!").expect("Should encode");
    // Encoding adds ▁ prefix to " hello" -> "▁hello" matches token 0
    assert!(!tokens.is_empty());
}

#[test]
fn test_phase33_loader_encode_no_vocab() {
    let data = build_gguf_header(0, 0);
    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert!(model.encode("hello").is_none());
}