burn-train 0.22.0-pre.1

Training crate for the Burn framework
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

use crate::learner::base::Interrupter;
use crate::metric::processor::{EventProcessorTraining, LearnerEvent, TrainingItem};
use crate::{
    InferenceStep, Learner, LearnerModel, SupervisedTrainingEventProcessor, TrainLoader,
    ValidLoader,
};
use burn_core::data::dataloader::Progress;
use burn_optim::GradientsAccumulator;

/// A validation epoch.
#[derive(new)]
pub struct SingleDeviceValidEpoch<M: LearnerModel> {
    dataloader: ValidLoader<M>,
}

/// A training epoch.
#[derive(new)]
pub struct SingleDeviceTrainEpoch<M: LearnerModel> {
    dataloader: TrainLoader<M>,
    grad_accumulation: Option<usize>,
}

impl<M: LearnerModel> SingleDeviceValidEpoch<M> {
    /// Runs the validation epoch.
    ///
    /// # Arguments
    ///
    /// * `model` - The model to validate.
    /// * `processor` - The event processor to use.
    pub fn run(
        &self,
        learner: &Learner<M>,
        global_progress: &Progress,
        processor: &mut SupervisedTrainingEventProcessor<M>,
        interrupter: &Interrupter,
    ) {
        let epoch = global_progress.items_processed;
        log::info!("Executing validation step for epoch {}", epoch);
        let model = learner.model().valid();

        let mut iterator = self.dataloader.iter();
        let mut iteration = 0;

        while let Some(item) = iterator.next() {
            let item = match item {
                Ok(item) => item,
                Err(err) => {
                    interrupter.stop(Some(&format!("dataset error during validation: {err}")));
                    break;
                }
            };
            let progress = iterator.progress();
            iteration += 1;

            let item = InferenceStep::step(&model, item);
            let item = TrainingItem::new(item, progress, Some(iteration), None);

            processor.process_valid(LearnerEvent::ProcessedItem(item));

            if interrupter.should_stop() {
                break;
            }
        }
    }
}

impl<M: LearnerModel> SingleDeviceTrainEpoch<M> {
    /// Runs the training epoch.
    ///
    /// # Arguments
    ///
    /// * `model` - The model to train.
    /// * `optim` - The optimizer to use.
    /// * `scheduler` - The learning rate scheduler to use.
    /// * `processor` - The event processor to use.
    ///
    /// # Returns
    ///
    /// The trained model and the optimizer.
    pub fn run(
        &self,
        learner: &mut Learner<M>,
        global_progress: &Progress,
        processor: &mut SupervisedTrainingEventProcessor<M>,
        interrupter: &Interrupter,
    ) {
        let epoch = global_progress.items_processed;
        log::info!("Executing training step for epoch {}", epoch,);

        // Single device / dataloader
        let mut iterator = self.dataloader.iter();
        let mut iteration = 0;
        let mut accumulator = GradientsAccumulator::new();
        let mut accumulation_current = 0;

        while let Some(item) = iterator.next() {
            let item = match item {
                Ok(item) => item,
                Err(err) => {
                    interrupter.stop(Some(&format!("dataset error during training: {err}")));
                    break;
                }
            };
            iteration += 1;
            learner.lr_step();
            log::info!("Iteration {iteration}");

            let progress = iterator.progress();
            let item = learner.train_step(item);

            match self.grad_accumulation {
                Some(accumulation) => {
                    accumulator.accumulate(&learner.model(), item.grads);
                    accumulation_current += 1;

                    if accumulation <= accumulation_current {
                        let grads = accumulator.grads();

                        learner.optimizer_step(grads);
                        accumulation_current = 0;
                    }
                }
                None => learner.optimizer_step(item.grads),
            }

            let item = TrainingItem::new(
                item.item,
                progress,
                Some(iteration),
                Some(learner.lr_current()),
            );

            processor.process_train(LearnerEvent::ProcessedItem(item));

            if interrupter.should_stop() {
                break;
            }
        }
    }
}