vectus 0.1.38

A vector database implemented in Rust for learning purposes.
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

#![allow(unused)]

pub mod document;
pub(crate) mod hnsw;
pub(crate) mod model;
pub use hnsw::metric;

use document::{DocBuilder, Document};
use hnsw::{metric::Metric, HNSWInitializer, HNSW};
use model::{Model, ModelType};
use ndarray::{Array1, Array2};
use std::{fmt::format, sync::Arc};
use tokio::sync::Mutex;

#[derive(Debug)]
pub enum StorageType {
    InMemory,
    Persistent,
}

pub struct Vectus {
    pub embeddings: Arc<Mutex<Array2<f32>>>,
    pub documents: Arc<Mutex<Vec<Document>>>,
    storage_type: StorageType,
    hnsw: Arc<Mutex<HNSW>>,
}

impl Vectus {
    pub fn new(storage_type: StorageType, metric: Metric) -> Vectus {
        let initializer = HNSWInitializer {
            max_level: 12,
            ef_construction: 350,
            m: 32,
            m_max: 64,
            norm: 3.0,
            entry: None,
            metric,
        };

        Vectus {
            embeddings: Mutex::new(Array2::zeros((0, 0))).into(),
            documents: Mutex::new(Vec::new()).into(),
            storage_type,
            hnsw: Mutex::new(HNSW::new(initializer)).into(),
        }
    }

    pub async fn get_k_relevant_documents(
        &self,
        api_key: &String,
        query: &String,
        k: usize,
    ) -> Vec<Document> {
        if query.is_empty() {
            return Vec::new();
        }

        let model = Model::new(ModelType::OpenAI(api_key.to_string()));

        let query_embedding = match model.get_embedding(query).await {
            Ok(embedding) => embedding,
            Err(e) => panic!("Error getting embedding: {}", e),
        };

        let query_emb = Array1::from_vec(query_embedding.clone());
        let hnsw_guard = self.hnsw.lock().await;

        let result = hnsw_guard
            .search(query_emb.clone(), hnsw_guard.len(), k)
            .await;

        let docs_guard = self.documents.lock().await;
        let mut relevant_docs: Vec<Document> = Vec::new();

        for ix in 0..k.min(result.len()) {
            relevant_docs.push(docs_guard[result[ix]].clone());
        }

        drop(hnsw_guard);
        drop(docs_guard);

        relevant_docs
    }

    pub async fn add_documents(
        &mut self,
        api_key: &String,
        docs: &Vec<Document>,
    ) -> Result<(), String> {
        if docs.is_empty() {
            return Err("No documents to add!".to_string());
        }

        let mut embeddings: Vec<Vec<f32>> = Vec::new();
        let model = Model::new(ModelType::OpenAI(api_key.to_string()));

        for doc in docs {
            let mut document = doc.page_content.clone();

            for (k, v) in doc.metadata.iter() {
                document.push_str(&format!("\n{} {}", k, v));
            }

            let embedding: Vec<f32> = match model.get_embedding(&document).await {
                Ok(embedding) => embedding,
                Err(e) => panic!("Error getting embedding: {}", e),
            };

            let nembd = Array1::from_vec(embedding.clone());
            self.store_emb_db(&nembd).await;
            embeddings.push(embedding);
        }

        let mut docs_guard = self.documents.lock().await;
        for doc in docs {
            docs_guard.push(doc.clone());
        }
        drop(docs_guard);

        let mut embeddings_guard = self.embeddings.lock().await;
        *embeddings_guard =
            Array2::from_shape_vec((docs.len(), embeddings[0].len()), embeddings.concat()).unwrap();

        Ok(())
    }

    async fn store_emb_db(&self, embedding: &Array1<f32>) {
        match self.storage_type {
            StorageType::InMemory => {
                let mut hnsw_guard = self.hnsw.lock().await;
                let len = hnsw_guard.len();
                hnsw_guard.insert(&embedding, len).await;
                drop(hnsw_guard);
            }
            StorageType::Persistent => {
                panic!("{:?} Not implemented yet!", self.storage_type);
            }
        }
    }
}