paddler_types 3.1.2

Open-source LLMOps platform for hosting and scaling AI in your own infrastructure
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

use anyhow::Result;
use anyhow::anyhow;
use serde::Deserialize;
use serde::Serialize;

use crate::embedding_normalization_method::EmbeddingNormalizationMethod;
use crate::normalization::l2;
use crate::normalization::rms_norm;
use crate::pooling_type::PoolingType;

#[derive(Debug, Deserialize, Serialize)]
#[serde(deny_unknown_fields)]
pub struct Embedding {
    pub embedding: Vec<f32>,
    pub normalization_method: EmbeddingNormalizationMethod,
    pub pooling_type: PoolingType,
    pub source_document_id: String,
}

impl Embedding {
    pub fn normalize(self, normalization_method: &EmbeddingNormalizationMethod) -> Result<Self> {
        if !self
            .normalization_method
            .can_transform_to(normalization_method)
        {
            return Err(anyhow!(
                "Cannot transform from {:?} to {normalization_method:?}",
                self.normalization_method
            ));
        }

        if !self
            .normalization_method
            .needs_transformation_to(normalization_method)
        {
            return Ok(self);
        }

        Ok(Self {
            embedding: match normalization_method {
                EmbeddingNormalizationMethod::None => self.embedding,
                EmbeddingNormalizationMethod::L2 => l2(&self.embedding),
                EmbeddingNormalizationMethod::RmsNorm { epsilon } => {
                    rms_norm(&self.embedding, *epsilon)
                }
            },
            normalization_method: normalization_method.clone(),
            pooling_type: self.pooling_type.clone(),
            source_document_id: self.source_document_id,
        })
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn make_embedding(values: Vec<f32>, method: EmbeddingNormalizationMethod) -> Embedding {
        Embedding {
            embedding: values,
            normalization_method: method,
            pooling_type: PoolingType::Mean,
            source_document_id: "test".to_owned(),
        }
    }

    #[test]
    fn test_normalize_from_none_to_l2() -> Result<()> {
        let embedding = make_embedding(vec![3.0, 4.0], EmbeddingNormalizationMethod::None);
        let result = embedding.normalize(&EmbeddingNormalizationMethod::L2)?;

        assert_eq!(result.embedding, vec![0.6, 0.8]);
        assert!(matches!(
            result.normalization_method,
            EmbeddingNormalizationMethod::L2
        ));

        Ok(())
    }

    #[test]
    fn test_normalize_from_none_to_rms_norm() -> Result<()> {
        let embedding =
            make_embedding(vec![2.0, 2.0, 2.0, 2.0], EmbeddingNormalizationMethod::None);
        let result =
            embedding.normalize(&EmbeddingNormalizationMethod::RmsNorm { epsilon: 0.0 })?;

        for val in &result.embedding {
            assert!((val - 1.0).abs() < 1e-6);
        }

        Ok(())
    }

    #[test]
    fn test_normalize_none_to_none_is_noop() -> Result<()> {
        let embedding = make_embedding(vec![1.0, 2.0, 3.0], EmbeddingNormalizationMethod::None);
        let result = embedding.normalize(&EmbeddingNormalizationMethod::None)?;

        assert_eq!(result.embedding, vec![1.0, 2.0, 3.0]);

        Ok(())
    }

    #[test]
    fn test_normalize_rejects_l2_to_rms_norm() {
        let embedding = make_embedding(vec![0.6, 0.8], EmbeddingNormalizationMethod::L2);
        let result = embedding.normalize(&EmbeddingNormalizationMethod::RmsNorm { epsilon: 1e-6 });

        assert!(result.is_err());
    }

    #[test]
    fn test_normalize_rejects_l2_to_none() {
        let embedding = make_embedding(vec![0.6, 0.8], EmbeddingNormalizationMethod::L2);
        let result = embedding.normalize(&EmbeddingNormalizationMethod::None);

        assert!(result.is_err());
    }

    #[test]
    fn test_normalize_rejects_rms_norm_to_l2() {
        let embedding = make_embedding(
            vec![1.0, 1.0],
            EmbeddingNormalizationMethod::RmsNorm { epsilon: 1e-6 },
        );
        let result = embedding.normalize(&EmbeddingNormalizationMethod::L2);

        assert!(result.is_err());
    }

    #[test]
    fn test_normalize_preserves_metadata() -> Result<()> {
        let embedding = Embedding {
            embedding: vec![3.0, 4.0],
            normalization_method: EmbeddingNormalizationMethod::None,
            pooling_type: PoolingType::Cls,
            source_document_id: "doc-42".to_owned(),
        };
        let result = embedding.normalize(&EmbeddingNormalizationMethod::L2)?;

        assert!(matches!(result.pooling_type, PoolingType::Cls));
        assert_eq!(result.source_document_id, "doc-42");

        Ok(())
    }
}