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


#[test]
fn test_phase35_array_of_bools() {
    let mut data = build_gguf_header(0, 1);
    let mut array_bytes = Vec::new();
    array_bytes.extend_from_slice(&7u32.to_le_bytes()); // bool type
    array_bytes.extend_from_slice(&3u64.to_le_bytes());
    array_bytes.push(1); // true
    array_bytes.push(0); // false
    array_bytes.push(1); // true
    data.extend(build_gguf_metadata("test_bools", 9, &array_bytes));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    if let Some(GGUFValue::Array(arr)) = model.metadata.get("test_bools") {
        assert_eq!(arr.len(), 3);
        assert!(matches!(arr[0], GGUFValue::Bool(true)));
        assert!(matches!(arr[1], GGUFValue::Bool(false)));
        assert!(matches!(arr[2], GGUFValue::Bool(true)));
    } else {
        panic!("Expected Array");
    }
}

#[test]
fn test_phase35_array_of_i64() {
    let mut data = build_gguf_header(0, 1);
    let mut array_bytes = Vec::new();
    array_bytes.extend_from_slice(&11u32.to_le_bytes()); // i64 type
    array_bytes.extend_from_slice(&2u64.to_le_bytes());
    array_bytes.extend_from_slice(&i64::MIN.to_le_bytes());
    array_bytes.extend_from_slice(&i64::MAX.to_le_bytes());
    data.extend(build_gguf_metadata("test_i64s", 9, &array_bytes));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    if let Some(GGUFValue::Array(arr)) = model.metadata.get("test_i64s") {
        assert_eq!(arr.len(), 2);
        assert!(matches!(arr[0], GGUFValue::Int64(v) if v == i64::MIN));
        assert!(matches!(arr[1], GGUFValue::Int64(v) if v == i64::MAX));
    } else {
        panic!("Expected Array");
    }
}

// =============================================================================
// Unicode and Special Characters
// =============================================================================

#[test]
fn test_phase35_unicode_tensor_names() {
    let mut data = build_gguf_header(1, 0);
    // Unicode tensor name with emojis and special chars
    data.extend(build_tensor_info("tensor_\u{1F4A1}_light", &[32], 0, 0));

    let current_len = data.len();
    let aligned = current_len.div_ceil(32) * 32;
    data.resize(aligned, 0);
    data.extend([0u8; 128]); // tensor data

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.tensors[0].name, "tensor_\u{1F4A1}_light");
}

#[test]
fn test_phase35_unicode_metadata_values() {
    let mut data = build_gguf_header(0, 1);
    let value = build_gguf_string("\u{4E2D}\u{6587}"); // Chinese characters
    data.extend(build_gguf_metadata("chinese_text", 8, &value));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    if let Some(GGUFValue::String(s)) = model.metadata.get("chinese_text") {
        assert_eq!(s, "\u{4E2D}\u{6587}");
    }
}

#[test]
fn test_phase35_long_string_metadata() {
    let mut data = build_gguf_header(0, 1);
    // 1KB string
    let long_value = "x".repeat(1024);
    let value = build_gguf_string(&long_value);
    data.extend(build_gguf_metadata("long_string", 8, &value));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    if let Some(GGUFValue::String(s)) = model.metadata.get("long_string") {
        assert_eq!(s.len(), 1024);
    }
}

// =============================================================================
// Multi-Tensor and Complex Models
// =============================================================================

#[test]
fn test_phase35_model_with_many_tensors() {
    let mut data = build_gguf_header(50, 0);

    // Add 50 tensor infos
    for i in 0..50 {
        data.extend(build_tensor_info(
            &format!("tensor_{:02}", i),
            &[32],
            0,
            (i * 128) as u64,
        ));
    }

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.tensors.len(), 50);
    assert_eq!(model.tensors[0].name, "tensor_00");
    assert_eq!(model.tensors[49].name, "tensor_49");
}

#[test]
fn test_phase35_model_with_many_metadata() {
    let mut data = build_gguf_header(0, 100);

    // Add 100 metadata entries
    for i in 0..100 {
        data.extend(build_gguf_metadata(
            &format!("key_{:03}", i),
            4,
            &(i as u32).to_le_bytes(),
        ));
    }

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    assert_eq!(model.metadata.len(), 100);
    assert!(matches!(
        model.metadata.get("key_000"),
        Some(GGUFValue::UInt32(0))
    ));
    assert!(matches!(
        model.metadata.get("key_099"),
        Some(GGUFValue::UInt32(99))
    ));
}

#[test]
fn test_phase35_full_model_config() {
    let mut data = build_gguf_header(2, 10);

    // Full set of metadata
    let arch = build_gguf_string("llama");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch));
    data.extend(build_gguf_metadata(
        "llama.embedding_length",
        4,
        &256u32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "llama.block_count",
        4,
        &4u32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "llama.attention.head_count",
        4,
        &8u32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "llama.attention.head_count_kv",
        4,
        &2u32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "llama.context_length",
        4,
        &2048u32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "llama.rope.freq_base",
        6,
        &10000.0f32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "llama.attention.layer_norm_rms_epsilon",
        6,
        &1e-5f32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.bos_token_id",
        4,
        &1u32.to_le_bytes(),
    ));
    data.extend(build_gguf_metadata(
        "tokenizer.ggml.eos_token_id",
        4,
        &2u32.to_le_bytes(),
    ));

    // Two tensors
    data.extend(build_tensor_info("token_embd.weight", &[100, 256], 0, 0));
    data.extend(build_tensor_info("output_norm.weight", &[256], 0, 102400));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");

    // Verify all accessors
    assert_eq!(model.architecture(), Some("llama"));
    assert_eq!(model.embedding_dim(), Some(256));
    assert_eq!(model.num_layers(), Some(4));
    assert_eq!(model.num_heads(), Some(8));
    assert_eq!(model.num_kv_heads(), Some(2));
    assert_eq!(model.context_length(), Some(2048));
    assert!((model.rope_freq_base().expect("test value should be present") - 10000.0).abs() < 1.0);
    assert!((model.rms_epsilon().expect("test value should be present") - 1e-5).abs() < 1e-7);
    assert_eq!(model.bos_token_id(), Some(1));
    assert_eq!(model.eos_token_id(), Some(2));
    assert_eq!(model.rope_type(), Some(0)); // NORM for llama
}

// =============================================================================
// GGUFModel Clone and Debug
// =============================================================================

#[test]
fn test_phase35_gguf_model_debug() {
    let data = build_gguf_header(0, 0);
    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let debug_str = format!("{:?}", model);
    assert!(debug_str.contains("GGUFModel"));
    assert!(debug_str.contains("header"));
}

#[test]
fn test_phase35_gguf_model_clone() {
    let mut data = build_gguf_header(1, 1);
    let arch = build_gguf_string("test");
    data.extend(build_gguf_metadata("general.architecture", 8, &arch));
    data.extend(build_tensor_info("test", &[32], 0, 0));

    let model = GGUFModel::from_bytes(&data).expect("Should parse");
    let cloned = model.clone();

    assert_eq!(model.header, cloned.header);
    assert_eq!(model.metadata.len(), cloned.metadata.len());
    assert_eq!(model.tensors.len(), cloned.tensors.len());
}