lib.rs 4.25 KB
Newer Older
1
2
3
mod app;
mod event;
mod generation;
4
mod table;
5
6
7
8
9
10
mod utils;

use crate::app::App;
use crate::event::Event;
use crossterm::ExecutableCommand;
use std::io;
11
use text_generation_client::{NextTokenChooserParameters, ShardedClient};
12
13
14
15
16
17
18
19
20
21
22
23
24
use tokenizers::Tokenizer;
use tokio::sync::{broadcast, mpsc};
use tui::backend::CrosstermBackend;
use tui::Terminal;

/// Run benchmarking app
#[allow(clippy::too_many_arguments)]
pub async fn run(
    tokenizer_name: String,
    tokenizer: Tokenizer,
    batch_size: Vec<u32>,
    sequence_length: u32,
    decode_length: u32,
Nicolas Patry's avatar
Nicolas Patry committed
25
    top_n_tokens: Option<u32>,
26
27
    n_runs: usize,
    warmups: usize,
28
29
30
31
32
    temperature: Option<f32>,
    top_k: Option<u32>,
    top_p: Option<f32>,
    typical_p: Option<f32>,
    repetition_penalty: Option<f32>,
33
    frequency_penalty: Option<f32>,
34
35
    watermark: bool,
    do_sample: bool,
36
    client: ShardedClient,
Nicolas Patry's avatar
Nicolas Patry committed
37
) -> Result<(), std::io::Error> {
38
39
40
41
42
43
44
45
    let parameters = NextTokenChooserParameters {
        temperature: temperature.unwrap_or(1.0),
        top_k: top_k.unwrap_or(0),
        top_p: top_p.unwrap_or(1.0),
        typical_p: typical_p.unwrap_or(1.0),
        do_sample,
        seed: 0,
        repetition_penalty: repetition_penalty.unwrap_or(1.0),
46
        frequency_penalty: frequency_penalty.unwrap_or(0.0),
47
48
49
        watermark,
    };

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
    // Initialize terminal properties
    crossterm::terminal::enable_raw_mode()?;
    io::stdout().execute(crossterm::terminal::EnterAlternateScreen)?;
    io::stdout().execute(crossterm::cursor::Hide)?;

    // Initialize terminal
    let mut terminal = {
        let backend = CrosstermBackend::new(io::stdout());
        Terminal::new(backend)?
    };

    // Create message channel between generation_task and app
    let (run_sender, run_receiver) = mpsc::channel(8);
    // Crossterm event channel
    let (event_sender, mut event_receiver) = mpsc::channel(8);
    // Shutdown channel to terminate tasks
    let (shutdown_sender, _) = broadcast::channel(1);
    // Channel to check if tasks terminated
    let (shutdown_guard_sender, mut shutdown_guard_receiver) = mpsc::channel(1);

    // Create generation task
    tokio::spawn(generation::generation_task(
        tokenizer,
        batch_size.clone(),
        sequence_length,
        decode_length,
Nicolas Patry's avatar
Nicolas Patry committed
76
        top_n_tokens,
77
78
        n_runs,
        warmups,
79
        parameters,
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
        client,
        run_sender,
        shutdown_sender.subscribe(),
        shutdown_guard_sender.clone(),
    ));

    // Create event task
    tokio::spawn(event::terminal_event_task(
        250,
        event_sender,
        shutdown_sender.subscribe(),
        shutdown_guard_sender.clone(),
    ));

    // Drop our end of shutdown sender
    drop(shutdown_guard_sender);

    // Create App
    let mut app = App::new(
        run_receiver,
100
        tokenizer_name.clone(),
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
        sequence_length,
        decode_length,
        n_runs,
        batch_size,
    );

    while app.running {
        // Draw frame
        terminal.draw(|frame| app.render(frame))?;

        // Await a new event from event handling task
        match event_receiver.recv().await {
            None => break,
            // Update app state
            Some(event) => match event {
                Event::Tick => app.tick(),
                Event::Key(key_event) => app.handle_key_event(key_event),
                _ => {}
            },
        }
    }

    // Ask tasks to shutdown
    let _ = shutdown_sender.send(());
    // Wait for tasks to shutdown
    let _ = shutdown_guard_receiver.recv().await;

    // Revert terminal to original view
    io::stdout().execute(crossterm::terminal::LeaveAlternateScreen)?;
    crossterm::terminal::disable_raw_mode()?;
    io::stdout().execute(crossterm::cursor::Show)?;

133
134
135
136
    let parameters_table = table::parameters_table(
        tokenizer_name,
        sequence_length,
        decode_length,
Nicolas Patry's avatar
Nicolas Patry committed
137
        top_n_tokens,
138
139
140
141
142
143
144
        n_runs,
        warmups,
        temperature,
        top_k,
        top_p,
        typical_p,
        repetition_penalty,
145
        frequency_penalty,
146
147
148
149
150
151
152
153
154
155
156
        watermark,
        do_sample,
    );
    println!("\n{parameters_table}\n");

    let latency_table = table::latency_table(&app.data);
    println!("\n{latency_table}\n");

    let throughput_table = table::throughput_table(&app.data);
    println!("\n{throughput_table}\n");

157
158
    Ok(())
}