embedrs 0.4.0

Unified embedding — cloud APIs (OpenAI, Cohere, Gemini, Voyage, Jina, Mistral) + local inference, one interface
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

#![cfg(feature = "local")]

//! End-to-end tests for local inference pipeline.
//!
//! These tests download real models from HuggingFace Hub on first run (~23-133MB each).
//! Run with: `cargo test -p embedrs --features local --test local_e2e -- --ignored`

use embedrs::Client;

fn l2_norm(v: &[f32]) -> f32 {
    v.iter().map(|x| x * x).sum::<f32>().sqrt()
}

fn cosine_similarity(a: &[f32], b: &[f32]) -> f32 {
    let dot: f32 = a.iter().zip(b.iter()).map(|(x, y)| x * y).sum();
    let norm_a = l2_norm(a);
    let norm_b = l2_norm(b);
    dot / (norm_a * norm_b)
}

#[tokio::test]
#[ignore]
async fn test_minilm_l6_embed_single() {
    let client = Client::local("all-MiniLM-L6-v2").unwrap();
    let result = client.embed(vec!["hello world".into()]).await.unwrap();

    assert_eq!(result.embeddings.len(), 1);
    assert_eq!(result.embeddings[0].len(), 384);
    let norm = l2_norm(&result.embeddings[0]);
    assert!(
        (norm - 1.0).abs() < 1e-4,
        "expected L2 norm ≈ 1.0, got {norm}"
    );
    assert!(
        result.usage.total_tokens > 0,
        "expected total_tokens > 0, got {}",
        result.usage.total_tokens
    );
}

#[tokio::test]
#[ignore]
async fn test_minilm_l12_embed() {
    let client = Client::local("all-MiniLM-L12-v2").unwrap();
    let result = client.embed(vec!["hello world".into()]).await.unwrap();

    assert_eq!(result.embeddings.len(), 1);
    assert_eq!(result.embeddings[0].len(), 384);
    let norm = l2_norm(&result.embeddings[0]);
    assert!(
        (norm - 1.0).abs() < 1e-4,
        "expected L2 norm ≈ 1.0, got {norm}"
    );
    assert!(
        result.usage.total_tokens > 0,
        "expected total_tokens > 0, got {}",
        result.usage.total_tokens
    );
}

#[tokio::test]
#[ignore]
async fn test_bge_small_embed() {
    let client = Client::local("bge-small-en-v1.5").unwrap();
    let result = client.embed(vec!["hello world".into()]).await.unwrap();

    assert_eq!(result.embeddings.len(), 1);
    assert_eq!(result.embeddings[0].len(), 384);
    let norm = l2_norm(&result.embeddings[0]);
    assert!(
        (norm - 1.0).abs() < 1e-4,
        "expected L2 norm ≈ 1.0, got {norm}"
    );
    assert!(
        result.usage.total_tokens > 0,
        "expected total_tokens > 0, got {}",
        result.usage.total_tokens
    );
}

#[tokio::test]
#[ignore]
async fn test_gte_small_embed() {
    let client = Client::local("gte-small").unwrap();
    let result = client.embed(vec!["hello world".into()]).await.unwrap();

    assert_eq!(result.embeddings.len(), 1);
    assert_eq!(result.embeddings[0].len(), 384);
    let norm = l2_norm(&result.embeddings[0]);
    assert!(
        (norm - 1.0).abs() < 1e-4,
        "expected L2 norm ≈ 1.0, got {norm}"
    );
    assert!(
        result.usage.total_tokens > 0,
        "expected total_tokens > 0, got {}",
        result.usage.total_tokens
    );
}

#[tokio::test]
#[ignore]
async fn test_semantic_similarity() {
    let client = Client::local("all-MiniLM-L6-v2").unwrap();
    let result = client
        .embed(vec![
            "dogs are great pets".into(),
            "cats are wonderful companions".into(),
            "quantum physics is complex".into(),
        ])
        .await
        .unwrap();

    assert_eq!(result.embeddings.len(), 3);

    let sim_dogs_cats = cosine_similarity(&result.embeddings[0], &result.embeddings[1]);
    let sim_dogs_quantum = cosine_similarity(&result.embeddings[0], &result.embeddings[2]);

    assert!(
        sim_dogs_cats > sim_dogs_quantum,
        "expected cosine(dogs, cats) > cosine(dogs, quantum), got {sim_dogs_cats} vs {sim_dogs_quantum}"
    );
}

#[tokio::test]
#[ignore]
async fn test_batch_local() {
    let client = Client::local("all-MiniLM-L6-v2").unwrap();
    let texts: Vec<String> = (0..20).map(|i| format!("sentence number {i}")).collect();
    let result = client.embed(texts).await.unwrap();

    assert_eq!(result.embeddings.len(), 20);
    for (i, emb) in result.embeddings.iter().enumerate() {
        assert_eq!(emb.len(), 384, "embedding {i} has wrong dimension");
        let norm = l2_norm(emb);
        assert!(
            (norm - 1.0).abs() < 1e-4,
            "embedding {i} L2 norm = {norm}, expected ≈ 1.0"
        );
    }
}

#[tokio::test]
#[ignore]
async fn test_deterministic() {
    let client = Client::local("all-MiniLM-L6-v2").unwrap();
    let text = vec!["determinism check".to_string()];

    let result1 = client.embed(text.clone()).await.unwrap();
    let result2 = client.embed(text).await.unwrap();

    assert_eq!(
        result1.embeddings[0], result2.embeddings[0],
        "expected identical embeddings for the same input"
    );
}

#[tokio::test]
#[ignore]
async fn test_empty_text() {
    let client = Client::local("all-MiniLM-L6-v2").unwrap();
    let result = client.embed(vec!["".to_string()]).await.unwrap();

    assert_eq!(result.embeddings.len(), 1);
    assert_eq!(result.embeddings[0].len(), 384);
}