a3s-code-core 8.0.3

A3S Code Core - Embeddable AI agent library with tool execution
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

use super::WorkspaceChunk;
use crate::embedding::{EmbeddingExecutorConfig, EmbeddingInput};
use std::sync::Arc;

#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub(super) enum SemanticBatchFlushReason {
    InputLimit,
    TextByteLimit,
    VectorByteLimit,
    GenerationComplete,
}

pub(super) struct SemanticBatchEntry {
    pub(super) path: String,
    pub(super) input: EmbeddingInput,
}

pub(super) struct SemanticBatch {
    pub(super) entries: Vec<SemanticBatchEntry>,
    pub(super) flush_reason: SemanticBatchFlushReason,
}

pub(super) struct SemanticBatchPlan {
    pub(super) batches: Vec<SemanticBatch>,
    pub(super) document_inputs: usize,
    pub(super) document_text_bytes: usize,
    pub(super) batch_limit_lower_bound: usize,
}

pub(super) fn plan_semantic_batches<'a>(
    partitions: impl IntoIterator<Item = (&'a str, &'a [Arc<WorkspaceChunk>])>,
    dimension: usize,
    config: EmbeddingExecutorConfig,
) -> SemanticBatchPlan {
    let vector_bytes_per_input = dimension.saturating_mul(std::mem::size_of::<f32>());
    let mut batches = Vec::new();
    let mut current = Vec::new();
    let mut current_text_bytes = 0usize;
    let mut document_inputs = 0usize;
    let mut document_text_bytes = 0usize;

    for (path, chunks) in partitions {
        for chunk in chunks {
            let input =
                EmbeddingInput::new(Arc::<str>::from(chunk.id.as_str()), Arc::clone(&chunk.text));
            document_inputs = document_inputs.saturating_add(1);
            document_text_bytes = document_text_bytes.saturating_add(input.text_bytes());

            if let Some(reason) = exceeded_reason(
                current.len(),
                current_text_bytes,
                input.text_bytes(),
                vector_bytes_per_input,
                config,
            ) {
                flush(&mut batches, &mut current, reason);
                current_text_bytes = 0;
            }

            current_text_bytes = current_text_bytes.saturating_add(input.text_bytes());
            current.push(SemanticBatchEntry {
                path: path.to_owned(),
                input,
            });

            if let Some(reason) = reached_reason(
                current.len(),
                current_text_bytes,
                vector_bytes_per_input,
                config,
            ) {
                flush(&mut batches, &mut current, reason);
                current_text_bytes = 0;
            }
        }
    }

    flush(
        &mut batches,
        &mut current,
        SemanticBatchFlushReason::GenerationComplete,
    );
    let vector_bytes = vector_bytes_per_input.saturating_mul(document_inputs);
    SemanticBatchPlan {
        batches,
        document_inputs,
        document_text_bytes,
        batch_limit_lower_bound: document_inputs
            .div_ceil(config.max_batch_inputs)
            .max(document_text_bytes.div_ceil(config.max_batch_text_bytes))
            .max(vector_bytes.div_ceil(config.max_batch_vector_bytes)),
    }
}

fn exceeded_reason(
    current_inputs: usize,
    current_text_bytes: usize,
    next_text_bytes: usize,
    vector_bytes_per_input: usize,
    config: EmbeddingExecutorConfig,
) -> Option<SemanticBatchFlushReason> {
    if current_inputs == 0 {
        return None;
    }
    if current_inputs.saturating_add(1) > config.max_batch_inputs {
        return Some(SemanticBatchFlushReason::InputLimit);
    }
    if current_text_bytes.saturating_add(next_text_bytes) > config.max_batch_text_bytes {
        return Some(SemanticBatchFlushReason::TextByteLimit);
    }
    if vector_bytes_per_input.saturating_mul(current_inputs.saturating_add(1))
        > config.max_batch_vector_bytes
    {
        return Some(SemanticBatchFlushReason::VectorByteLimit);
    }
    None
}

fn reached_reason(
    current_inputs: usize,
    current_text_bytes: usize,
    vector_bytes_per_input: usize,
    config: EmbeddingExecutorConfig,
) -> Option<SemanticBatchFlushReason> {
    if current_inputs == config.max_batch_inputs {
        return Some(SemanticBatchFlushReason::InputLimit);
    }
    if current_text_bytes == config.max_batch_text_bytes {
        return Some(SemanticBatchFlushReason::TextByteLimit);
    }
    if vector_bytes_per_input.saturating_mul(current_inputs) == config.max_batch_vector_bytes {
        return Some(SemanticBatchFlushReason::VectorByteLimit);
    }
    None
}

fn flush(
    batches: &mut Vec<SemanticBatch>,
    current: &mut Vec<SemanticBatchEntry>,
    flush_reason: SemanticBatchFlushReason,
) {
    if current.is_empty() {
        return;
    }
    batches.push(SemanticBatch {
        entries: std::mem::take(current),
        flush_reason,
    });
}