lattice-embed 0.15.0

SIMD-accelerated vector operations and embedding generation
Documentation
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
422
423
424
425
426
427
428
use super::*;
use crate::model::EmbeddingModel;

#[test]
fn test_cache_basic_operations() {
    let cache = EmbeddingCache::new(100);
    let key = cache.compute_key(
        "hello",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );

    assert!(cache.get(&key).is_none());

    let embedding = vec![0.1, 0.2, 0.3];
    cache.put(key, embedding.clone());

    let cached = cache.get(&key).unwrap();
    assert_eq!(&*cached, &embedding[..]);
}

#[test]
fn test_cache_eviction() {
    // With 16 shards, a capacity of 16 gives 1 entry per shard.
    // To test eviction, we need keys that hash to the same shard.
    // Use a larger capacity and fill it up.
    let cache = EmbeddingCache::new(16);

    // Insert 32 entries — each shard has capacity 1, so each shard
    // can only hold 1 entry. Inserting 2 entries to the same shard
    // will evict the first.
    let mut keys = Vec::new();
    for i in 0..32u32 {
        let text = format!("text_{i}");
        let key = cache.compute_key(
            &text,
            ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
            EmbeddingRole::Generic,
        );
        keys.push(key);
        cache.put(key, vec![i as f32]);
    }

    // Total size should not exceed capacity (16)
    let stats = cache.stats();
    assert!(stats.size <= 16, "size {} exceeds capacity 16", stats.size);
}

#[test]
fn test_cache_lru_eviction_within_shard() {
    // Create cache with capacity 32 (2 per shard).
    let cache = EmbeddingCache::new(32);

    // Find 3 keys that land in the same shard.
    let mut same_shard_keys = Vec::new();
    let mut i = 0u32;

    // Find the first key's shard and collect 3 keys for it.
    let first_key = cache.compute_key(
        "probe_0",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    let target_shard = shard_index(&first_key);

    loop {
        let key = cache.compute_key(
            &format!("lru_test_{i}"),
            ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
            EmbeddingRole::Generic,
        );
        if shard_index(&key) == target_shard {
            same_shard_keys.push((key, i));
        }
        if same_shard_keys.len() == 3 {
            break;
        }
        i += 1;
    }

    let (k1, v1) = same_shard_keys[0];
    let (k2, v2) = same_shard_keys[1];
    let (k3, v3) = same_shard_keys[2];

    // Insert k1 and k2 (shard capacity is 2).
    cache.put(k1, vec![v1 as f32]);
    cache.put(k2, vec![v2 as f32]);

    // Access k1 to make it recently used.
    assert!(cache.get(&k1).is_some());

    // Insert k3 — should evict k2 (least recently used in this shard).
    cache.put(k3, vec![v3 as f32]);

    assert!(
        cache.get(&k1).is_some(),
        "k1 should survive (recently accessed)"
    );
    assert!(cache.get(&k2).is_none(), "k2 should be evicted (LRU)");
    assert!(cache.get(&k3).is_some(), "k3 should exist (just inserted)");
}

#[test]
fn test_cache_different_models_different_keys() {
    let cache = EmbeddingCache::new(100);

    let key_small = cache.compute_key(
        "text",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    let key_base = cache.compute_key(
        "text",
        ModelConfig::new(EmbeddingModel::BgeBaseEnV15),
        EmbeddingRole::Generic,
    );

    // Same text, different models = different keys
    assert_ne!(key_small, key_base);
}

#[test]
fn test_cache_stats() {
    let cache = EmbeddingCache::new(100);
    let key = cache.compute_key(
        "hello",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );

    cache.get(&key); // Miss
    cache.put(key, vec![0.1]);
    cache.get(&key); // Hit

    let stats = cache.stats();
    assert_eq!(stats.size, 1);
    assert_eq!(stats.hits, 1);
    assert_eq!(stats.misses, 1);
    assert!((stats.hit_rate() - 0.5).abs() < 0.001);
}

#[test]
fn test_cache_get_many() {
    // Use capacity large enough that no shard evicts.
    let cache = EmbeddingCache::new(100);

    let key1 = cache.compute_key(
        "one",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    let key2 = cache.compute_key(
        "two",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    let key3 = cache.compute_key(
        "three",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );

    cache.put(key1, vec![1.0]);
    cache.put(key3, vec![3.0]);

    let results = cache.get_many(&[key1, key2, key3]);
    assert_eq!(results.len(), 3);
    assert_eq!(&**results[0].as_ref().unwrap(), &[1.0f32]);
    assert!(results[1].is_none());
    assert_eq!(&**results[2].as_ref().unwrap(), &[3.0f32]);
}

#[test]
fn test_cache_put_many() {
    // Use capacity large enough that no shard evicts (ceil(100/16) = 7 per shard).
    let cache = EmbeddingCache::new(100);

    let key1 = cache.compute_key(
        "one",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    let key2 = cache.compute_key(
        "two",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );

    cache.put_many(vec![(key1, vec![1.0]), (key2, vec![2.0])]);

    let v1 = cache.get(&key1).unwrap();
    assert_eq!(&*v1, [1.0f32].as_slice());
    let v2 = cache.get(&key2).unwrap();
    assert_eq!(&*v2, [2.0f32].as_slice());
}

#[test]
fn test_cache_clear() {
    let cache = EmbeddingCache::new(100);
    let key = cache.compute_key(
        "hello",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );

    cache.put(key, vec![0.1]);
    assert!(cache.get(&key).is_some());

    cache.clear();
    assert!(cache.get(&key).is_none());
    assert_eq!(cache.stats().size, 0);
}

#[test]
fn test_cache_default_capacity() {
    let cache = EmbeddingCache::with_default_capacity();
    assert_eq!(cache.stats().capacity, DEFAULT_CACHE_CAPACITY);
}

#[test]
fn test_cache_disabled_is_noop() {
    let cache = EmbeddingCache::new(0);
    assert!(!cache.is_enabled());

    let key = cache.compute_key(
        "hello",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    cache.put(key, vec![0.1]);
    assert!(cache.get(&key).is_none());

    let stats = cache.stats();
    assert_eq!(stats.capacity, 0);
    assert_eq!(stats.size, 0);
}

#[test]
fn test_concurrent_access() {
    use std::thread;

    // Use large capacity so no eviction occurs (800 entries across 16 shards).
    // Per-shard capacity = ceil(4000/16) = 250, so 800 entries fit easily.
    let cache = Arc::new(EmbeddingCache::new(4000));
    let mut handles = Vec::new();

    // Spawn 8 threads, each doing 100 put+get operations.
    for t in 0..8 {
        let cache = Arc::clone(&cache);
        handles.push(thread::spawn(move || {
            for i in 0..100 {
                // Each thread uses unique keys to avoid contention on same entry.
                let text = format!("thread_{t}_item_{i}");
                let key = cache.compute_key(
                    &text,
                    ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
                    EmbeddingRole::Generic,
                );
                let embedding = vec![t as f32; 384];
                cache.put(key, embedding.clone());

                let result = cache.get(&key);
                assert!(result.is_some(), "put followed by get must succeed");
                assert_eq!(result.unwrap().len(), 384);
            }
        }));
    }

    for h in handles {
        h.join().expect("thread panicked");
    }

    // All 800 entries should be in cache (capacity 4000 >> 800).
    let stats = cache.stats();
    assert_eq!(stats.size, 800);
    assert!(stats.hits >= 800, "at least 800 hits expected");
}

#[test]
fn test_shard_distribution() {
    // Use generous capacity to avoid eviction from uneven distribution.
    // 4000 total → 250/shard. We insert 800 entries → ~50/shard on average.
    let cache = EmbeddingCache::new(4000);

    let n = 800;
    for i in 0..n {
        let key = cache.compute_key(
            &format!("item_{i}"),
            ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
            EmbeddingRole::Generic,
        );
        cache.put(key, vec![i as f32]);
    }

    let shard_stats = cache.per_shard_stats();
    assert_eq!(shard_stats.len(), NUM_SHARDS);

    // Each shard should have entries. With uniform hash, each shard gets ~50.
    for ss in &shard_stats {
        assert!(
            ss.size > 0,
            "shard {} is empty — distribution is pathological",
            ss.shard_id
        );
    }

    // No eviction since 800 << 4000. Total should be exactly 800.
    let total: usize = shard_stats.iter().map(|s| s.size).sum();
    assert_eq!(total, n);

    // Check distribution is reasonably uniform: no shard has >3x the average.
    let avg = n / NUM_SHARDS; // 50
    for ss in &shard_stats {
        assert!(
            ss.size <= avg * 3,
            "shard {} has {} entries (avg {}), distribution too skewed",
            ss.shard_id,
            ss.size,
            avg
        );
    }
}

#[test]
fn test_per_shard_stats_hit_tracking() {
    let cache = EmbeddingCache::new(100);

    // Insert a few entries and access them.
    let key1 = cache.compute_key(
        "hello",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    let key2 = cache.compute_key(
        "world",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );

    cache.put(key1, vec![1.0]);
    cache.put(key2, vec![2.0]);

    // Access key1 three times, key2 once.
    cache.get(&key1);
    cache.get(&key1);
    cache.get(&key1);
    cache.get(&key2);

    let shard_stats = cache.per_shard_stats();
    let total_hits: u64 = shard_stats.iter().map(|s| s.hits).sum();
    assert_eq!(total_hits, 4, "total hits should be 4");

    let stats = cache.stats();
    assert_eq!(stats.hits, 4);
    assert_eq!(stats.misses, 0);
}

#[test]
fn test_small_capacity_rounds_up() {
    // Capacity smaller than NUM_SHARDS: each shard gets at least 1.
    let cache = EmbeddingCache::new(3);
    assert!(cache.is_enabled());

    let key = cache.compute_key(
        "x",
        ModelConfig::new(EmbeddingModel::BgeSmallEnV15),
        EmbeddingRole::Generic,
    );
    cache.put(key, vec![42.0]);
    assert!(cache.get(&key).is_some());
}

// -------------------------------------------------------------------------
// Role-aware cache key tests (P0-E2)
// -------------------------------------------------------------------------

/// Query and passage roles must produce different cache keys for the same raw text
/// so that embed_query("hello") and embed_passage("hello") are stored separately.
#[test]
fn test_role_query_vs_passage_different_keys() {
    let cache = EmbeddingCache::new(100);
    let model = ModelConfig::new(EmbeddingModel::MultilingualE5Small);
    let text = "hello world";

    let key_query = cache.compute_key(text, model, EmbeddingRole::Query);
    let key_passage = cache.compute_key(text, model, EmbeddingRole::Passage);
    let key_generic = cache.compute_key(text, model, EmbeddingRole::Generic);

    assert_ne!(key_query, key_passage, "query vs passage must differ");
    assert_ne!(key_query, key_generic, "query vs generic must differ");
    assert_ne!(key_passage, key_generic, "passage vs generic must differ");
}

/// Role keys are consistent: same inputs always produce same key.
#[test]
fn test_role_key_deterministic() {
    let cache = EmbeddingCache::new(100);
    let model = ModelConfig::new(EmbeddingModel::BgeSmallEnV15);

    let k1 = cache.compute_key("test", model, EmbeddingRole::Query);
    let k2 = cache.compute_key("test", model, EmbeddingRole::Query);
    assert_eq!(k1, k2, "identical inputs must produce identical key");
}

/// Storing under one role does not pollute the other role's key.
#[test]
fn test_role_cache_isolation() {
    let cache = EmbeddingCache::new(100);
    let model = ModelConfig::new(EmbeddingModel::MultilingualE5Small);

    let key_query = cache.compute_key("embed me", model, EmbeddingRole::Query);
    let key_passage = cache.compute_key("embed me", model, EmbeddingRole::Passage);

    // Store under Query role.
    cache.put(key_query, vec![1.0, 2.0]);

    // Passage role key must still be a miss.
    assert!(
        cache.get(&key_passage).is_none(),
        "passage key must miss after storing under query key"
    );

    // Query role key must hit.
    assert!(
        cache.get(&key_query).is_some(),
        "query key must hit after storing under query key"
    );
}