aprender-rag 0.64.0

Pure-Rust Retrieval-Augmented Generation pipeline built on Trueno
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

#![allow(clippy::disallowed_methods)]
//! Benchmarks for retrieval operations

use aprender_rag::{
    chunk::{Chunk, Chunker, RecursiveChunker},
    embed::MockEmbedder,
    index::{BM25Index, SparseIndex, VectorStore},
    Document, DocumentId, Embedder,
};
use criterion::{criterion_group, criterion_main, Criterion};
use std::hint::black_box;

fn create_test_chunk(content: &str, embedding: Vec<f32>) -> Chunk {
    let mut chunk = Chunk::new(DocumentId::new(), content.to_string(), 0, content.len());
    chunk.set_embedding(embedding);
    chunk
}

fn bench_bm25_indexing(c: &mut Criterion) {
    let mut group = c.benchmark_group("bm25_indexing");

    let chunks: Vec<_> = (0..1000)
        .map(|i| {
            Chunk::new(
                DocumentId::new(),
                format!("Document {i} contains information about machine learning and artificial intelligence"),
                0,
                80,
            )
        })
        .collect();

    group.bench_function("index_1000_chunks", |b| {
        b.iter(|| {
            let mut index = BM25Index::new();
            for chunk in &chunks {
                index.add(black_box(chunk));
            }
            index
        });
    });

    group.finish();
}

fn bench_bm25_search(c: &mut Criterion) {
    let mut group = c.benchmark_group("bm25_search");

    // Pre-build index
    let mut index = BM25Index::new();
    for i in 0..1000 {
        let chunk = Chunk::new(
            DocumentId::new(),
            format!("Document {i} about topic {} with keywords", i % 100),
            0,
            50,
        );
        index.add(&chunk);
    }

    group.bench_function("search_top_10", |b| {
        b.iter(|| index.search(black_box("topic keywords"), 10));
    });

    group.bench_function("search_top_100", |b| {
        b.iter(|| index.search(black_box("topic keywords"), 100));
    });

    group.finish();
}

fn bench_vector_search(c: &mut Criterion) {
    let mut group = c.benchmark_group("vector_search");

    // Pre-build store
    let mut store = VectorStore::with_dimension(128);
    for i in 0..1000 {
        let mut embedding = vec![0.0f32; 128];
        embedding[i % 128] = 1.0;
        let chunk = create_test_chunk(&format!("document {i}"), embedding);
        store.insert(chunk).unwrap();
    }

    let query = vec![1.0f32; 128];

    group.bench_function("search_top_10", |b| {
        b.iter(|| store.search(black_box(&query), 10));
    });

    group.bench_function("search_top_100", |b| {
        b.iter(|| store.search(black_box(&query), 100));
    });

    group.finish();
}

fn bench_chunking(c: &mut Criterion) {
    let mut group = c.benchmark_group("chunking");

    let long_doc = Document::new("Lorem ipsum dolor sit amet. ".repeat(1000));
    let chunker = RecursiveChunker::new(512, 50);

    group.bench_function("chunk_large_doc", |b| {
        b.iter(|| chunker.chunk(black_box(&long_doc)));
    });

    group.finish();
}

fn bench_embedding(c: &mut Criterion) {
    let mut group = c.benchmark_group("embedding");

    let embedder = MockEmbedder::new(384);
    let texts: Vec<&str> = (0..100).map(|_| "This is a test sentence for embedding").collect();

    group.bench_function("embed_100_texts", |b| {
        b.iter(|| embedder.embed_batch(black_box(&texts)));
    });

    group.finish();
}

criterion_group!(
    benches,
    bench_bm25_indexing,
    bench_bm25_search,
    bench_vector_search,
    bench_chunking,
    bench_embedding,
);

criterion_main!(benches);