mullama 0.3.0

Comprehensive Rust bindings for llama.cpp with memory-safe API and advanced features
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
150
151
152
153
154
155
156
157
158

use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;

use tokio::sync::mpsc;

use super::super::super::protocol::{ResponseFormat, StreamChunk};
use super::super::prompt::find_stop_in_recent_window;
use crate::{Context, Model, MullamaError, SamplerChain};

/// Controls whether tokens are buffered or streamed.
pub(super) enum TokenSink<'a> {
    Buffer,
    Stream {
        tx: &'a mpsc::Sender<StreamChunk>,
        request_id: Arc<str>,
        cancel_flag: &'a AtomicBool,
    },
}

/// Result of the core token generation loop.
pub(super) struct GenerationResult {
    pub generated: String,
    pub completion_tokens: u32,
}

/// Core token generation loop shared by all generation paths.
pub(super) fn generate_tokens(
    context: &mut Context,
    model: &Model,
    sampler: &mut SamplerChain,
    max_tokens: u32,
    stop_sequences: &[String],
    max_stop_len: usize,
    sink: &TokenSink<'_>,
) -> Result<GenerationResult, MullamaError> {
    let mut generated = String::with_capacity(match sink {
        TokenSink::Buffer => (max_tokens as usize) * 6,
        TokenSink::Stream { .. } => 256,
    });
    let mut completion_tokens = 0u32;

    let mut index = 0u32;
    let mut sent_len = 0usize;
    let hold_back = max_stop_len.saturating_sub(1);
    let mut last_token_id = 0i32;

    for _ in 0..max_tokens {
        if let TokenSink::Stream { cancel_flag, .. } = sink {
            if cancel_flag.load(Ordering::Relaxed) {
                break;
            }
        }

        let next_token = sampler.sample(context, -1);

        if model.vocab_is_eog(next_token) {
            break;
        }

        if let Ok(text) = model.token_to_str(next_token, 0, false) {
            let previous_len = generated.len();
            generated.push_str(&text);

            if let Some(pos) =
                find_stop_in_recent_window(&generated, previous_len, stop_sequences, max_stop_len)
            {
                if let TokenSink::Stream { tx, request_id, .. } = sink {
                    if pos > sent_len {
                        let chunk = StreamChunk {
                            request_id: request_id.clone(),
                            index,
                            delta: generated[sent_len..pos].to_string(),
                            token_id: next_token,
                            thinking: None,
                            tool_calls: None,
                        };
                        let _ = tx.blocking_send(chunk);
                    }
                }
                generated.truncate(pos);
                return Ok(GenerationResult {
                    generated,
                    completion_tokens,
                });
            }

            if let TokenSink::Stream { tx, request_id, .. } = sink {
                let mut flush_end = generated.len().saturating_sub(hold_back);
                while flush_end > sent_len && !generated.is_char_boundary(flush_end) {
                    flush_end -= 1;
                }
                if flush_end > sent_len {
                    let chunk = StreamChunk {
                        request_id: request_id.clone(),
                        index,
                        delta: generated[sent_len..flush_end].to_string(),
                        token_id: next_token,
                        thinking: None,
                        tool_calls: None,
                    };
                    if tx.blocking_send(chunk).is_err() {
                        break;
                    }
                    sent_len = flush_end;
                    index += 1;
                }
            }

            last_token_id = next_token;
        }

        sampler.accept(next_token);
        context.decode_single(next_token)?;
        completion_tokens += 1;
    }

    if let TokenSink::Stream { tx, request_id, .. } = sink {
        if sent_len < generated.len() {
            let chunk = StreamChunk {
                request_id: request_id.clone(),
                index,
                delta: generated[sent_len..].to_string(),
                token_id: last_token_id,
                thinking: None,
                tool_calls: None,
            };
            let _ = tx.blocking_send(chunk);
        }
    }

    Ok(GenerationResult {
        generated,
        completion_tokens,
    })
}

/// Resolve a response format into an optional GBNF grammar string.
pub(super) fn resolve_grammar(response_format: Option<&ResponseFormat>) -> Option<String> {
    match response_format {
        Some(ResponseFormat::JsonSchema { json_schema }) => {
            match crate::structured_output::JsonSchemaConverter::convert(&json_schema.schema) {
                Ok(grammar) => Some(grammar.to_gbnf()),
                Err(e) => {
                    tracing::warn!("Failed to convert JSON schema to grammar: {}", e);
                    None
                }
            }
        }
        Some(ResponseFormat::JsonObject) => match crate::grammar::presets::json() {
            Ok(grammar) => Some(grammar.to_gbnf()),
            Err(e) => {
                tracing::warn!("Failed to create JSON grammar: {}", e);
                None
            }
        },
        Some(ResponseFormat::Text) | None => None,
    }
}