diffai-core 0.5.1

Core library for AI/ML diff analysis - PyTorch, Safetensors, tensor statistics
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

use serde_json::Value;

/// Analyze epoch progression patterns
pub(crate) fn analyze_epoch_progression(
    old_obj: &serde_json::Map<String, Value>,
    new_obj: &serde_json::Map<String, Value>,
) -> Option<(String, String)> {
    let old_epoch = extract_epoch_info(old_obj)?;
    let new_epoch = extract_epoch_info(new_obj)?;

    if new_epoch <= old_epoch {
        return None; // No progression or regression
    }

    let epoch_diff = new_epoch - old_epoch;
    let progression_rate = if epoch_diff == 1.0 {
        "normal"
    } else if epoch_diff < 1.0 {
        "fractional"
    } else {
        "skipped_epochs"
    };

    let old_info = format!("epoch: {old_epoch}");
    let new_info =
        format!("epoch: {new_epoch}, progression: {progression_rate} ({epoch_diff:+.1})");

    Some((old_info, new_info))
}

/// Extract epoch information from model checkpoint
pub(crate) fn extract_epoch_info(obj: &serde_json::Map<String, Value>) -> Option<f64> {
    if let Some(Value::Number(num)) = obj.get("epoch") {
        return num.as_f64();
    }
    None
}