realizar 0.8.5

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


#[test]
fn test_pipeline_parallel_invalid_stage() {
    let result = PipelineParallel::new(4, 10, 24, 4);
    assert!(result.is_err());
}

#[test]
fn test_pipeline_parallel_bubble_ratio_zero_microbatches() {
    let pp = PipelineParallel::new(4, 0, 24, 4).expect("test");
    let ratio = pp.bubble_ratio(0);
    assert_eq!(ratio, 1.0);
}

#[test]
fn test_pipeline_parallel_bubble_ratio_many_microbatches() {
    let pp = PipelineParallel::new(4, 0, 24, 4).expect("test");
    // With many micro-batches, bubble ratio approaches 0
    let ratio = pp.bubble_ratio(1000);
    assert!(ratio < 0.01);
}

#[test]
fn test_pipeline_parallel_record_multiple_microbatches() {
    let mut pp = PipelineParallel::new(4, 0, 24, 4).expect("test");

    // Record several micro-batches
    pp.record_micro_batch(10.0);
    pp.record_micro_batch(20.0);
    pp.record_micro_batch(30.0);

    let stats = pp.stats();
    assert_eq!(stats.micro_batches_processed, 3);
    assert_eq!(stats.forward_passes, 3);
    // Running average: (10 + 20 + 30) / 3 = 20
    assert!((stats.avg_stage_latency_ms - 20.0).abs() < 0.1);
}

#[test]
fn test_pipeline_stage_clone_serialize() {
    let stage = PipelineStage {
        index: 1,
        start_layer: 6,
        end_layer: 12,
        num_layers: 6,
    };
    let cloned = stage.clone();
    assert_eq!(cloned.index, 1);

    let json = serde_json::to_string(&stage).expect("serialize");
    let deserialized: PipelineStage = serde_json::from_str(&json).expect("deserialize");
    assert_eq!(deserialized.num_layers, 6);
}

// =========================================================================
// Additional Coverage Tests: ParallelConfig Ranks
// =========================================================================

#[test]
fn test_parallel_config_is_tp_last() {
    // tp_size=2, last TP rank is 1
    let config = ParallelConfig::new(2, 1, 1, 1).expect("test");
    assert!(config.is_tp_last());
    assert!(!config.is_tp_first());
}

#[test]
fn test_parallel_config_is_pp_last() {
    // pp_size=2, last PP stage is at rank 1 (with tp_size=1)
    let config = ParallelConfig::new(1, 2, 1, 1).expect("test");
    assert!(config.is_pp_last());
    assert!(!config.is_pp_first());
}

#[test]
fn test_parallel_config_complex_ranks() {
    // World: tp=2, pp=3, dp=2 => 12 ranks
    // Rank 7: tp_rank = 7 % 2 = 1, pp_stage = (7/2) % 3 = 0, dp_rank = 7/(2*3) = 1
    let config = ParallelConfig::new(2, 3, 2, 7).expect("test");
    assert_eq!(config.tp_rank(), 1);
    assert_eq!(config.pp_stage(), 0);
    assert_eq!(config.dp_rank(), 1);
}

#[test]
fn test_parallel_config_default() {
    let config = ParallelConfig::default();
    assert_eq!(config.tp_size, 1);
    assert_eq!(config.pp_size, 1);
    assert_eq!(config.dp_size, 1);
    assert_eq!(config.rank, 0);
}

#[test]
fn test_parallel_config_serialization() {
    let config = ParallelConfig::new(2, 2, 2, 3).expect("test");
    let json = serde_json::to_string(&config).expect("serialize");
    let deserialized: ParallelConfig = serde_json::from_str(&json).expect("deserialize");
    assert_eq!(deserialized.tp_size, 2);
    assert_eq!(deserialized.rank, 3);
}

// =========================================================================
// Additional Coverage Tests: DistributedContext
// =========================================================================

#[test]
fn test_distributed_context_pipeline_parallel_mut() {
    let config = ParallelConfig::new(1, 4, 1, 0).expect("test");
    let mut ctx = DistributedContext::new(config).expect("test");

    ctx.init_pipeline(24, 4).expect("test");

    // Use pipeline_parallel_mut
    if let Some(pp) = ctx.pipeline_parallel_mut() {
        pp.record_micro_batch(15.0);
    }

    let stats = ctx.pipeline_parallel().expect("pp").stats();
    assert_eq!(stats.micro_batches_processed, 1);
}

#[test]
fn test_distributed_context_config_accessor() {
    let config = ParallelConfig::new(2, 2, 2, 5).expect("test");
    let ctx = DistributedContext::new(config).expect("test");

    let cfg = ctx.config();
    assert_eq!(cfg.tp_size, 2);
    assert_eq!(cfg.rank, 5);
}

#[test]
fn test_distributed_context_init_pipeline_single_pp() {
    // When pp_size = 1, init_pipeline should not create PipelineParallel
    let config = ParallelConfig::new(2, 1, 1, 0).expect("test");
    let mut ctx = DistributedContext::new(config).expect("test");

    ctx.init_pipeline(24, 4).expect("test");
    assert!(ctx.pipeline_parallel().is_none());
}

#[test]
fn test_distributed_context_debug() {
    let config = ParallelConfig::single();
    let ctx = DistributedContext::new(config).expect("test");
    let debug_str = format!("{:?}", ctx);
    assert!(debug_str.contains("config"));
    assert!(debug_str.contains("initialized"));
}

// =========================================================================
// Additional Coverage Tests: ZeroOffload Serialization
// =========================================================================

#[test]
fn test_zero_offload_serialization() {
    let zero = ZeroOffload::inference();
    let json = serde_json::to_string(&zero).expect("serialize");
    let deserialized: ZeroOffload = serde_json::from_str(&json).expect("deserialize");
    assert!(deserialized.offload_params);
    assert!(deserialized.offload_activations);
}

// =========================================================================
// Additional Coverage Tests: ParallelError NotInitialized
// =========================================================================

#[test]
fn test_parallel_error_not_initialized_display() {
    let err = ParallelError::NotInitialized;
    let msg = err.to_string();
    assert!(msg.contains("not initialized"));
}

#[test]
fn test_parallel_error_pipeline_error_display() {
    let err = ParallelError::PipelineError("stage mismatch".to_string());
    let msg = err.to_string();
    assert!(msg.contains("stage mismatch"));
}

#[test]
fn test_parallel_error_invalid_world_size_display() {
    let err = ParallelError::InvalidWorldSize(0);
    let msg = err.to_string();
    assert!(msg.contains("0"));
}

// =========================================================================
// Additional Coverage Tests: ParallelTensor Operations
// =========================================================================

#[test]
fn test_parallel_tensor_zeros_various_shapes() {
    let t1 = ParallelTensor::zeros(vec![1]);
    assert_eq!(t1.numel(), 1);

    let t2 = ParallelTensor::zeros(vec![2, 3, 4]);
    assert_eq!(t2.numel(), 24);
    assert_eq!(t2.sum(), 0.0);
}

#[test]
fn test_parallel_tensor_sum_negative_values() {
    let tensor = ParallelTensor::new(vec![4], vec![-1.0, 2.0, -3.0, 4.0]).expect("test");
    assert_eq!(tensor.sum(), 2.0);
}

#[test]
#[allow(clippy::many_single_char_names)]
fn test_parallel_tensor_large_matmul() {
    // Larger matrix multiplication
    let m = 4;
    let k = 8;
    let n = 4;
    let a = ParallelTensor::new(vec![m, k], vec![1.0; m * k]).expect("test");
    let b = ParallelTensor::new(vec![k, n], vec![1.0; k * n]).expect("test");
    let c = a.matmul(&b).expect("test");

    assert_eq!(c.shape, vec![m, n]);
    // Each element should be k (sum of 1.0 * 1.0 k times)
    assert_eq!(c.data[0], k as f32);
}

#[test]
fn test_parallel_tensor_clone() {
    let tensor = ParallelTensor::new(vec![2, 2], vec![1.0, 2.0, 3.0, 4.0]).expect("test");
    let cloned = tensor.clone();
    assert_eq!(cloned.shape, tensor.shape);
    assert_eq!(cloned.data, tensor.data);
}

#[test]
fn test_parallel_tensor_debug() {
    let tensor = ParallelTensor::new(vec![2], vec![1.0, 2.0]).expect("test");
    let debug_str = format!("{:?}", tensor);
    assert!(debug_str.contains("shape"));
    assert!(debug_str.contains("data"));
}

// =========================================================================
// Additional Coverage Tests: PipelineStats Default
// =========================================================================

#[test]
fn test_pipeline_stats_default() {
    let stats = PipelineStats::default();
    assert_eq!(stats.micro_batches_processed, 0);
    assert_eq!(stats.forward_passes, 0);
    assert_eq!(stats.bubble_time_ms, 0.0);
    assert_eq!(stats.avg_stage_latency_ms, 0.0);
}

#[test]
fn test_pipeline_stats_serialization() {
    let stats = PipelineStats {
        micro_batches_processed: 50,
        forward_passes: 50,
        bubble_time_ms: 2.5,
        avg_stage_latency_ms: 8.0,
    };
    let json = serde_json::to_string(&stats).expect("serialize");
    let deserialized: PipelineStats = serde_json::from_str(&json).expect("deserialize");
    assert_eq!(deserialized.micro_batches_processed, 50);
    assert!((deserialized.avg_stage_latency_ms - 8.0).abs() < 0.01);
}