chonkier 0.0.2

🦛 Chonkie, now in Rust 🦀: No-nonsense, ultra-fast, ultra-light chunking library
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

/*
   Token Chunker chunks the given string into chunks based on the
   a tokenizer. It takes in a tokenizer, with chunk_size and chunk_overlap
   and returns a list of chunks.
*/

use crate::tokenizer::{CharacterTokenizer, Tokenizer};
use crate::types::Chunk;
use rayon::prelude::*;

pub struct TokenChunker<T: Tokenizer> {
    pub tokenizer: T,
    pub chunk_size: usize,
    pub chunk_overlap: usize,
}

impl<T: Tokenizer> TokenChunker<T> {
    pub fn new(tokenizer: T, chunk_size: usize, chunk_overlap: usize) -> Self {
        Self {
            tokenizer,
            chunk_size,
            chunk_overlap,
        }
    }

    pub fn chunk(&self, text: &str) -> Vec<Chunk> {
        // Encode the text into tokens
        let tokens = self.tokenizer.encode(text);

        // Pre-allocate with capacity to avoid reallocations
        let mut chunks = Vec::with_capacity(tokens.len().div_ceil(self.chunk_size));

        let mut current_pos = 0;
        while current_pos < tokens.len() {
            let end = (current_pos + self.chunk_size).min(tokens.len());
            let chunk_tokens = &tokens[current_pos..end];

            // Decode the current chunk
            let chunk_text = self.tokenizer.decode(chunk_tokens);

            // Find the position in the original text
            let start_index = if current_pos == 0 {
                0
            } else {
                // For subsequent chunks, we can use the previous chunk's end position
                chunks.last().map_or(0, |c: &Chunk| c.end_index)
            };

            let end_index = start_index + chunk_text.len();

            chunks.push(Chunk::new(
                chunk_text,
                start_index,
                end_index,
                chunk_tokens.len(),
            ));

            current_pos += self.chunk_size - self.chunk_overlap;
        }

        chunks
    }

    pub fn chunk_batch(&self, texts: &Vec<String>) -> Vec<Vec<Chunk>> {
        texts.par_iter().map(|text| self.chunk(text)).collect()
    }
}

impl Default for TokenChunker<CharacterTokenizer> {
    fn default() -> Self {
        Self {
            tokenizer: CharacterTokenizer::new(),
            chunk_size: 512,
            chunk_overlap: 0,
        }
    }
}

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

    #[test]
    fn test_chunk() {
        let tokenizer = CharacterTokenizer::new();
        let chunker = TokenChunker::new(tokenizer, 12, 0);
        let text = "Hello, world! This is a test.";
        let chunks = chunker.chunk(&text.to_string());
        assert_eq!(chunks.len(), 3);
        assert_eq!(chunks[0].text, "Hello, world");
        assert_eq!(chunks[0].start_index, 0);
        assert_eq!(chunks[0].end_index, 12);
    }

    #[test]
    fn test_chunk_batch() {
        let tokenizer = CharacterTokenizer::new();
        let chunker = TokenChunker::new(tokenizer, 12, 0);
        let texts = vec![
            "Hello, world! This is a test.".to_string(),
            "This is another test.".to_string(),
        ];
        let chunks = chunker.chunk_batch(&texts);
        assert_eq!(chunks.len(), 2);
        assert_eq!(chunks[0].len(), 3);
        assert_eq!(chunks[1].len(), 2);
    }
}