osf-rs 0.0.1

OSF Sleep Foundation Model — inference in Rust with Burn ML
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

/// Example: Extract embeddings from synthetic PSG data.
///
/// Usage:
///   cargo run --example embed --release -- --weights data/osf_backbone.safetensors

use std::path::Path;
use std::time::Instant;
use clap::Parser;

// ── Backend ───────────────────────────────────────────────────────────────────
#[cfg(feature = "wgpu")]
mod backend {
    pub use burn::backend::Wgpu as B;
    pub fn device() -> burn::backend::wgpu::WgpuDevice {
        burn::backend::wgpu::WgpuDevice::default()
    }
}
#[cfg(not(feature = "wgpu"))]
mod backend {
    pub use burn::backend::NdArray as B;
    pub fn device() -> burn::backend::ndarray::NdArrayDevice {
        burn::backend::ndarray::NdArrayDevice::Cpu
    }
}
use backend::B;

#[derive(Parser, Debug)]
#[command(about = "OSF — PSG embedding extraction")]
struct Args {
    #[arg(long)]
    weights: String,
    #[arg(long)]
    config: Option<String>,
    #[arg(long, default_value = "data/embeddings.safetensors")]
    output: String,
    #[arg(long, short = 'v')]
    verbose: bool,
}

fn generate_synthetic_psg(n_channels: usize, n_samples: usize) -> Vec<f32> {
    let mut signal = vec![0.0f32; n_channels * n_samples];
    for ch in 0..n_channels {
        let freq = 1.0 + ch as f32 * 0.5;
        let mut noise_state: u32 = (ch as u32 + 1) * 0xDEAD_BEEF;
        for t in 0..n_samples {
            let time = t as f32 / 64.0; // 64 Hz
            let sine = (2.0 * std::f32::consts::PI * freq * time).sin() * 50e-6;
            noise_state ^= noise_state << 13;
            noise_state ^= noise_state >> 17;
            noise_state ^= noise_state << 5;
            let noise = (noise_state as f32 / u32::MAX as f32 - 0.5) * 10e-6;
            signal[ch * n_samples + t] = sine + noise;
        }
    }
    signal
}

fn main() -> anyhow::Result<()> {
    let args = Args::parse();
    let t0 = Instant::now();
    let device = backend::device();

    println!("╔══════════════════════════════════════════════════════════════╗");
    println!("║  OSF — PSG Embedding Extraction                             ║");
    println!("╚══════════════════════════════════════════════════════════════╝\n");

    // 1. Load model
    let model_cfg = if let Some(ref cfg_path) = args.config {
        let cfg_str = std::fs::read_to_string(cfg_path)?;
        serde_json::from_str(&cfg_str)?
    } else {
        osf_rs::ModelConfig::default()
    };

    println!("▸ Loading OSF model …");
    let (encoder, ms_load) = osf_rs::OsfEncoder::<B>::load_with_config(
        model_cfg,
        Path::new(&args.weights),
        device.clone(),
    )?;
    println!("  {}  ({ms_load:.0} ms)\n", encoder.describe());

    // 2. Generate synthetic PSG epochs
    let n_channels = osf_rs::NUM_PSG_CHANNELS;
    let n_samples = osf_rs::EPOCH_SAMPLES;
    let n_epochs = 3;

    println!("▸ Generating {} synthetic PSG epochs ({} ch × {} samples each)\n",
        n_epochs, n_channels, n_samples);

    let mut all_outputs = Vec::new();

    for epoch_idx in 0..n_epochs {
        let signal = generate_synthetic_psg(n_channels, n_samples);
        let batch = osf_rs::build_batch::<B>(signal, n_channels, n_samples, &device);

        let t = Instant::now();
        let result = encoder.run_batch(&batch)?;
        let ms = t.elapsed().as_secs_f64() * 1000.0;

        // Stats
        let cls = &result.cls_emb;
        let mean: f64 = cls.iter().map(|&v| v as f64).sum::<f64>() / cls.len() as f64;
        let std: f64 = (cls.iter().map(|&v| { let d = v as f64 - mean; d * d }).sum::<f64>()
            / cls.len() as f64).sqrt();

        println!("  Epoch {epoch_idx}: cls=[{}]  patches=[{},{}]  mean={mean:+.4}  std={std:.4}  {ms:.1}ms",
            result.embed_dim, result.num_patches, result.embed_dim);

        if args.verbose && epoch_idx == 0 {
            println!("    First 5 CLS values: {:?}", &cls[..5.min(cls.len())]);
        }

        all_outputs.push(result);
    }

    // 3. Save
    let encoding = osf_rs::EncodingResult {
        epochs: all_outputs,
        ms_load,
        ms_encode: t0.elapsed().as_secs_f64() * 1000.0,
    };

    if let Some(p) = Path::new(&args.output).parent() {
        std::fs::create_dir_all(p)?;
    }
    encoding.save_safetensors(&args.output)?;
    println!("\n▸ Saved {} epochs → {}", n_epochs, args.output);

    let ms_total = t0.elapsed().as_secs_f64() * 1000.0;
    println!("\n  Total: {ms_total:.0} ms");
    Ok(())
}