burn_dragon_vision 0.4.0

Foveation and vision sampling utilities for burn dragon
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
150

#[cfg(feature = "cli")]
use crate::train::prelude::*;
#[cfg(feature = "cli")]
use crate::train::vision::train_vision_backend;
#[cfg(feature = "cli")]
use burn_dragon_language::train::train_backend;
#[cfg(feature = "cli")]
use burn_dragon_language::{load_training_config, train::build_vocab_only, train::prepare_dataset};

#[cfg(feature = "cli")]
fn run_in_training_thread<F, T>(name: &str, work: F) -> Result<T>
where
    F: FnOnce() -> Result<T> + Send + 'static,
    T: Send + 'static,
{
    #[cfg(target_os = "windows")]
    {
        let handle = std::thread::Builder::new()
            .name(name.to_string())
            .spawn(work)
            .context("spawn training thread")?;
        handle
            .join()
            .map_err(|_| anyhow!("training thread panicked"))?
    }
    #[cfg(not(target_os = "windows"))]
    {
        work()
    }
}

#[cfg(feature = "cli")]
#[derive(Parser, Debug)]
#[command(author, version, about = "Train the Baby Dragon Hatchling model")]
struct Cli {
    #[command(flatten)]
    train: TrainArgs,
    #[command(subcommand)]
    command: Option<Command>,
}

#[cfg(feature = "cli")]
#[derive(ClapArgs, Debug)]
struct TrainArgs {
    /// Additional configuration files applied in order (later files override earlier ones).
    #[arg(short = 'c', long = "config", value_name = "PATH", global = true)]
    config: Vec<PathBuf>,
    /// Backend to use for training.
    #[arg(long, value_enum, default_value_t = BackendArg::Cuda)]
    backend: BackendArg,
}

#[cfg(feature = "cli")]
#[derive(Subcommand, Debug)]
enum Command {
    /// Build the character-level vocabulary and exit.
    BuildVocab,
    /// Train the vision model (distill or LeJEPA).
    Vision,
}

#[cfg(feature = "cli")]
#[derive(Copy, Clone, Debug, ValueEnum)]
enum BackendArg {
    Cuda,
    Wgpu,
    WgpuNoFusion,
}

#[cfg(feature = "cli")]
pub fn run_cli() -> Result<()> {
    let args = Cli::parse();

    if matches!(args.command, Some(Command::Vision)) {
        let backend = args.train.backend;
        let mut config_paths = vec![PathBuf::from("config/vision/base.toml")];
        config_paths.extend(args.train.config.clone());
        let config = load_vision_training_config(&config_paths)?;
        return run_in_training_thread("vision-train", move || match backend {
            BackendArg::Wgpu => {
                train_vision_backend::<Autodiff<Wgpu<f32>>, _>(&config, "wgpu", |device| {
                    init_runtime(device, &config.wgpu)
                })
            }
            BackendArg::WgpuNoFusion => {
                use burn_wgpu::{CubeBackend, WgpuRuntime};
                type WgpuNoFusion = CubeBackend<WgpuRuntime, f32, i32, u32>;
                train_vision_backend::<Autodiff<WgpuNoFusion>, _>(&config, "wgpu-nofusion", |device| {
                    init_runtime(device, &config.wgpu)
                })
            }
            BackendArg::Cuda => {
                #[cfg(feature = "cuda")]
                {
                    train_vision_backend::<Autodiff<Cuda<f32>>, _>(&config, "cuda", |_| {})
                }
                #[cfg(not(feature = "cuda"))]
                {
                    Err(anyhow!(
                        "cuda backend selected but this build lacks `cuda` feature; rebuild with `--features cuda`"
                    ))
                }
            }
        });
    }

    let backend = args.train.backend;
    let mut config_paths = vec![PathBuf::from("config/language/base.toml")];
    config_paths.extend(args.train.config.clone());
    let config = load_training_config(&config_paths)?;
    FAST_TRAIN.store(config.training.fast_train, Ordering::Relaxed);

    if matches!(args.command, Some(Command::BuildVocab)) {
        build_vocab_only(&config)?;
        return Ok(());
    }

    let dataset = prepare_dataset(&config.dataset, &config.training)?;

    run_in_training_thread("train", move || match backend {
        BackendArg::Wgpu => train_backend::<Autodiff<Wgpu<f32>>, _>(
            &config,
            Arc::clone(&dataset),
            "wgpu",
            |device| init_runtime(device, &config.wgpu),
        ),
        BackendArg::WgpuNoFusion => {
            use burn_wgpu::{CubeBackend, WgpuRuntime};
            type WgpuNoFusion = CubeBackend<WgpuRuntime, f32, i32, u32>;
            train_backend::<Autodiff<WgpuNoFusion>, _>(
                &config,
                Arc::clone(&dataset),
                "wgpu-nofusion",
                |device| init_runtime(device, &config.wgpu),
            )
        }
        BackendArg::Cuda => {
            #[cfg(feature = "cuda")]
            {
                train_backend::<Autodiff<Cuda<f32>>, _>(&config, dataset, "cuda", |_| {})
            }
            #[cfg(not(feature = "cuda"))]
            {
                Err(anyhow!(
                    "cuda backend selected but this build lacks `cuda` feature; rebuild with `--features cuda`"
                ))
            }
        }
    })
}