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
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421

    // ==========================================================================
    // Test Helpers: Mock types for creating test APR model bytes
    // Note: These are test-only mocks; real APR inference is not yet implemented
    // ==========================================================================

    /// Mock model weights structure for test APR file creation
    /// This creates valid APR file format for testing the Lambda handler scaffolding
    #[derive(serde::Serialize)]
    struct MockModelWeights {
        weights: Vec<Vec<f32>>,
        biases: Vec<Vec<f32>>,
        dimensions: Vec<usize>,
    }

    /// Mock model type enum for test APR file creation
    #[repr(u16)]
    #[derive(Clone, Copy)]
    enum MockAprModelType {
        LinearRegression = 1,
    }

    impl MockAprModelType {
        fn as_u16(self) -> u16 {
            self as u16
        }
    }

    /// Create APR model bytes where output = sum of inputs (for easy verification)
    /// Note: Real inference is not implemented yet; tests verify handler scaffolding
    fn create_sum_model(input_dim: usize) -> &'static [u8] {
        // Single output that sums all inputs: output[0] = sum(input[i])
        let weights = MockModelWeights {
            weights: vec![vec![1.0; input_dim]], // 1 x input_dim weights (all 1.0)
            biases: vec![vec![0.0]],             // Single bias of 0
            dimensions: vec![input_dim, 1],
        };

        let payload = serde_json::to_vec(&weights).expect("serialize weights");
        let mut data = Vec::with_capacity(HEADER_SIZE + payload.len());

        data.extend_from_slice(&MAGIC);
        data.push(1);
        data.push(0);
        data.push(0);
        data.push(0);
        data.extend_from_slice(&MockAprModelType::LinearRegression.as_u16().to_le_bytes());
        data.extend_from_slice(&0u32.to_le_bytes());
        #[allow(clippy::cast_possible_truncation)]
        data.extend_from_slice(&(payload.len() as u32).to_le_bytes());
        #[allow(clippy::cast_possible_truncation)]
        data.extend_from_slice(&(payload.len() as u32).to_le_bytes());
        data.extend_from_slice(&[0u8; 10]);
        data.extend_from_slice(&payload);

        Box::leak(data.into_boxed_slice())
    }

    // ==========================================================================
    // GREEN PHASE: Tests for Lambda handler with real APR inference
    // ==========================================================================

    // --------------------------------------------------------------------------
    // Test: LambdaHandler creation
    // --------------------------------------------------------------------------

    #[test]
    fn test_lambda_handler_creation_with_valid_apr_model() {
        let model_bytes = create_sum_model(3);
        let handler = LambdaHandler::from_bytes(model_bytes);
        assert!(handler.is_ok(), "Should accept valid .apr model");
        let handler = handler.expect("test");
        assert!(handler.model_size_bytes() > HEADER_SIZE);
    }

    #[test]
    fn test_lambda_handler_rejects_empty_model() {
        let model_bytes: &'static [u8] = b"";
        let result = LambdaHandler::from_bytes(model_bytes);
        assert!(result.is_err());
        assert_eq!(result.unwrap_err(), LambdaError::EmptyModel);
    }

    #[test]
    fn test_lambda_handler_rejects_invalid_magic() {
        // Invalid magic bytes (GGUF instead of APR)
        let model_bytes: &'static [u8] = b"GGUF\x01\x00\x00\x00testmodel";
        let result = LambdaHandler::from_bytes(model_bytes);
        assert!(result.is_err());
        match result.unwrap_err() {
            LambdaError::InvalidMagic { expected, found } => {
                assert_eq!(expected, "APR");
                assert!(!found.is_empty());
                assert!(found.contains("71")); // 'G' = 71 in ASCII
            },
            _ => panic!("Expected InvalidMagic error"),
        }
    }

    // --------------------------------------------------------------------------
    // Test: Lambda request/response serialization
    // --------------------------------------------------------------------------

    #[test]
    fn test_lambda_request_serialization() {
        let request = LambdaRequest {
            features: vec![0.5, 1.2, -0.3, 0.8],
            model_id: Some("sentiment-v1".to_string()),
        };

        let json = serde_json::to_string(&request).expect("serialization failed");
        assert!(json.contains("0.5"));
        assert!(json.contains("sentiment-v1"));
    }

    #[test]
    fn test_lambda_request_deserialization() {
        let json = r#"{"features": [1.0, 2.0, 3.0], "model_id": "test-model"}"#;
        let request: LambdaRequest = serde_json::from_str(json).expect("deserialization failed");
        assert_eq!(request.features, vec![1.0, 2.0, 3.0]);
        assert_eq!(request.model_id, Some("test-model".to_string()));
    }

    #[test]
    fn test_lambda_response_serialization() {
        let response = LambdaResponse {
            prediction: 0.85,
            probabilities: Some(vec![0.15, 0.85]),
            latency_ms: 2.3,
            cold_start: true,
        };

        let json = serde_json::to_string(&response).expect("serialization failed");
        assert!(json.contains("0.85"));
        assert!(json.contains("cold_start"));
        assert!(json.contains("true"));
    }

    // --------------------------------------------------------------------------
    // Test: Lambda handler invocation
    // --------------------------------------------------------------------------

    #[test]
    #[ignore = "Mock APR format incompatible with real AprModel::from_bytes parser"]
    fn test_lambda_handler_cold_start_detection() {
        let model_bytes = create_sum_model(3);
        let handler = LambdaHandler::from_bytes(model_bytes).expect("test");

        // Before first invocation: cold start
        assert!(handler.is_cold_start());

        let request = LambdaRequest {
            features: vec![1.0, 2.0, 3.0],
            model_id: None,
        };

        // First invocation
        let response = handler.handle(&request).expect("test");
        assert!(response.cold_start, "First invocation should be cold start");

        // After first invocation: no longer cold
        assert!(!handler.is_cold_start());

        // Second invocation
        let response2 = handler.handle(&request).expect("test");
        assert!(
            !response2.cold_start,
            "Second invocation should not be cold start"
        );
    }

    #[test]
    fn test_lambda_handler_rejects_empty_features() {
        let model_bytes = create_sum_model(3);
        let handler = LambdaHandler::from_bytes(model_bytes).expect("test");

        let request = LambdaRequest {
            features: vec![],
            model_id: None,
        };

        let result = handler.handle(&request);
        assert!(result.is_err());
        assert_eq!(result.unwrap_err(), LambdaError::EmptyFeatures);
    }

    #[test]
    #[ignore = "Mock APR format incompatible with real AprModel::from_bytes parser"]
    fn test_lambda_handler_real_inference() {
        // Create a model where output = sum of 3 inputs
        let model_bytes = create_sum_model(3);
        let handler = LambdaHandler::from_bytes(model_bytes).expect("test");

        let request = LambdaRequest {
            features: vec![1.0, 2.0, 3.0],
            model_id: None,
        };

        let response = handler.handle(&request).expect("test");
        // Real inference: sum model returns sum of features
        // output = 1.0 * 1.0 + 1.0 * 2.0 + 1.0 * 3.0 = 6.0
        assert!((response.prediction - 6.0).abs() < 0.001);
        assert!(response.latency_ms >= 0.0);
    }

    // --------------------------------------------------------------------------
    // Test: Cold start metrics
    // --------------------------------------------------------------------------

    #[test]
    #[ignore = "Mock APR format incompatible with real AprModel::from_bytes parser"]
    fn test_cold_start_metrics_recorded() {
        let model_bytes = create_sum_model(1);
        let handler = LambdaHandler::from_bytes(model_bytes).expect("test");

        // No metrics before first invocation
        assert!(handler.cold_start_metrics().is_none());

        let request = LambdaRequest {
            features: vec![1.0],
            model_id: None,
        };

        let _ = handler.handle(&request).expect("test");

        // Metrics available after first invocation
        let metrics = handler.cold_start_metrics();
        assert!(metrics.is_some());
        let metrics = metrics.expect("test");
        assert!(metrics.total_ms >= 0.0);
        assert!(metrics.first_inference_ms >= 0.0);
        assert!(metrics.model_load_ms >= 0.0);
    }

    // --------------------------------------------------------------------------
    // Test: ARM64 optimizations
    // --------------------------------------------------------------------------

    #[test]
    fn test_arm64_architecture_detection() {
        let arch = arm64::target_arch();
        assert!(
            arch == "aarch64" || arch == "x86_64" || arch == "unknown",
            "Should detect valid architecture"
        );
    }

    #[test]
    fn test_arm64_simd_detection() {
        let simd = arm64::optimal_simd();
        let valid_simd = ["NEON", "AVX2", "SSE2", "Scalar"];
        assert!(
            valid_simd.contains(&simd),
            "Should detect valid SIMD backend"
        );
    }

    // --------------------------------------------------------------------------
    // Test: Benchmark utilities
    // --------------------------------------------------------------------------

    #[test]
    #[ignore = "Mock APR format incompatible with real AprModel::from_bytes parser"]
    fn test_benchmark_cold_start() {
        let model_bytes = create_sum_model(2);
        let handler = LambdaHandler::from_bytes(model_bytes).expect("test");

        let request = LambdaRequest {
            features: vec![1.0, 2.0],
            model_id: None,
        };

        let result = benchmark::benchmark_cold_start(&handler, &request, 10).expect("test");

        assert!(result.cold_start_ms >= 0.0);
        assert!(result.warm_inference_ms >= 0.0);
        assert_eq!(result.warm_iterations, 10);
        assert!(result.model_size_bytes > 0);
    }

    #[test]
    fn test_benchmark_targets() {
        // Verify target constants match spec §8.1
        assert!((benchmark::TARGET_COLD_START_MS - 50.0).abs() < f64::EPSILON);
        assert!((benchmark::TARGET_WARM_INFERENCE_MS - 10.0).abs() < f64::EPSILON);
    }

    #[test]
    fn test_benchmark_result_meets_targets() {
        let result = benchmark::BenchmarkResult {
            cold_start_ms: 45.0,
            warm_inference_ms: 8.0,
            warm_iterations: 100,
            model_size_bytes: 1000,
            target_arch: "aarch64".to_string(),
            simd_backend: "NEON".to_string(),
            meets_cold_start_target: true,
            meets_warm_inference_target: true,
        };

        assert!(result.meets_all_targets());
    }

    #[test]
    fn test_benchmark_result_fails_targets() {
        let result = benchmark::BenchmarkResult {
            cold_start_ms: 75.0, // Exceeds 50ms target
            warm_inference_ms: 8.0,
            warm_iterations: 100,
            model_size_bytes: 1000,
            target_arch: "aarch64".to_string(),
            simd_backend: "NEON".to_string(),
            meets_cold_start_target: false,
            meets_warm_inference_target: true,
        };

        assert!(!result.meets_all_targets());
    }

    // --------------------------------------------------------------------------
    // Test: Error display
    // --------------------------------------------------------------------------

    #[test]
    fn test_lambda_error_display() {
        assert_eq!(LambdaError::EmptyModel.to_string(), "Model bytes are empty");
        assert_eq!(
            LambdaError::EmptyFeatures.to_string(),
            "Request features are empty"
        );
        assert_eq!(
            LambdaError::EmptyBatch.to_string(),
            "Batch request has no instances"
        );
        assert!(LambdaError::InvalidMagic {
            expected: "APR\\0".to_string(),
            found: "GGUF".to_string()
        }
        .to_string()
        .contains("Invalid magic"));
    }

    // --------------------------------------------------------------------------
    // Test: Batch inference (PROD-001)
    // Per spec §5.3 and §11.3
    // --------------------------------------------------------------------------

    #[test]
    fn test_batch_request_serialization() {
        let request = BatchLambdaRequest {
            instances: vec![
                LambdaRequest {
                    features: vec![1.0, 2.0],
                    model_id: None,
                },
                LambdaRequest {
                    features: vec![3.0, 4.0],
                    model_id: None,
                },
            ],
            max_parallelism: Some(4),
        };

        let json = serde_json::to_string(&request).expect("serialization failed");
        assert!(json.contains("instances"));
        assert!(json.contains("max_parallelism"));
        assert!(json.contains("1.0"));
        assert!(json.contains("3.0"));
    }

    #[test]
    fn test_batch_response_serialization() {
        let response = BatchLambdaResponse {
            predictions: vec![LambdaResponse {
                prediction: 3.0,
                probabilities: None,
                latency_ms: 1.5,
                cold_start: false,
            }],
            total_latency_ms: 5.0,
            success_count: 1,
            error_count: 0,
        };

        let json = serde_json::to_string(&response).expect("serialization failed");
        assert!(json.contains("predictions"));
        assert!(json.contains("success_count"));
        assert!(json.contains("total_latency_ms"));
    }

    #[test]
    #[ignore = "Mock APR format incompatible with real AprModel::from_bytes parser"]
    fn test_batch_handler_success() {
        let model_bytes = create_sum_model(2);
        let handler = LambdaHandler::from_bytes(model_bytes).expect("test");

        let request = BatchLambdaRequest {
            instances: vec![
                LambdaRequest {
                    features: vec![1.0, 2.0],
                    model_id: None,
                },
                LambdaRequest {
                    features: vec![3.0, 4.0],
                    model_id: None,
                },
            ],
            max_parallelism: None,
        };

        let response = handler.handle_batch(&request).expect("test");

        assert_eq!(response.predictions.len(), 2);
        assert_eq!(response.success_count, 2);
        assert_eq!(response.error_count, 0);
        assert!(response.total_latency_ms >= 0.0);

        // Real inference: sum model returns sum of features
        assert!((response.predictions[0].prediction - 3.0).abs() < 0.001);
        assert!((response.predictions[1].prediction - 7.0).abs() < 0.001);
    }