burn-train 0.21.0-pre.5

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

use crate::metric::{AccuracyInput, PerplexityInput, TopKAccuracyInput};
use crate::metric::{Adaptor, CerInput, LossInput, WerInput, processor::ItemLazy};
use burn_core::tensor::backend::Backend;
use burn_core::tensor::{Int, Tensor, Transaction};
use burn_flex::Flex;

/// Sequence prediction output adapted for multiple metrics.
///
/// Supported metrics:
/// - Accuracy
/// - TopKAccuracy
/// - Perplexity
/// - Loss
/// - CER
/// - WER
#[derive(new)]
pub struct SequenceOutput<B: Backend> {
    /// The loss.
    pub loss: Tensor<B, 1>,

    /// Raw logits. Shape: `[batch_size, seq_len, vocab_size]`
    pub logits: Tensor<B, 3>,

    /// Optional predicted token indices. Shape: `[batch_size, seq_length]`.
    /// If not provided, predictions default to argmax of `logits` along the last dimension.
    pub predictions: Option<Tensor<B, 2, Int>>,

    /// The target token indices. Shape: `[batch_size, seq_length]`
    pub targets: Tensor<B, 2, Int>,
}

impl<B: Backend> SequenceOutput<B> {
    fn predicted_tokens(&self) -> Tensor<B, 2, Int> {
        match &self.predictions {
            Some(preds) => preds.clone(),
            None => self.logits.clone().argmax(2).squeeze_dim::<2>(2),
        }
    }

    fn flat_logits(&self) -> Tensor<B, 2> {
        let [batch_size, seq_len, vocab_size] = self.logits.dims();
        self.logits
            .clone()
            .reshape([batch_size * seq_len, vocab_size])
    }

    fn flat_targets(&self) -> Tensor<B, 1, Int> {
        let [batch_size, seq_len] = self.targets.dims();
        self.targets.clone().reshape([batch_size * seq_len])
    }
}

impl<B: Backend> ItemLazy for SequenceOutput<B> {
    // Flex's IntElem is i32; token IDs > i32::MAX would truncate on sync.
    type ItemSync = SequenceOutput<Flex>;

    fn sync(self) -> Self::ItemSync {
        let device = &Default::default();

        match self.predictions {
            Some(preds) => {
                let [logits, loss, targets, predictions] = Transaction::default()
                    .register(self.logits)
                    .register(self.loss)
                    .register(self.targets)
                    .register(preds)
                    .execute()
                    .try_into()
                    .expect("Correct amount of tensor data");

                SequenceOutput {
                    logits: Tensor::from_data(logits, device),
                    loss: Tensor::from_data(loss, device),
                    targets: Tensor::from_data(targets, device),
                    predictions: Some(Tensor::from_data(predictions, device)),
                }
            }
            None => {
                let [logits, loss, targets] = Transaction::default()
                    .register(self.logits)
                    .register(self.loss)
                    .register(self.targets)
                    .execute()
                    .try_into()
                    .expect("Correct amount of tensor data");

                SequenceOutput {
                    logits: Tensor::from_data(logits, device),
                    loss: Tensor::from_data(loss, device),
                    targets: Tensor::from_data(targets, device),
                    predictions: None,
                }
            }
        }
    }
}

impl<B: Backend> Adaptor<LossInput<B>> for SequenceOutput<B> {
    fn adapt(&self) -> LossInput<B> {
        LossInput::new(self.loss.clone())
    }
}

impl<B: Backend> Adaptor<CerInput<B>> for SequenceOutput<B> {
    fn adapt(&self) -> CerInput<B> {
        CerInput::new(self.predicted_tokens(), self.targets.clone())
    }
}

impl<B: Backend> Adaptor<WerInput<B>> for SequenceOutput<B> {
    fn adapt(&self) -> WerInput<B> {
        WerInput::new(self.predicted_tokens(), self.targets.clone())
    }
}

impl<B: Backend> Adaptor<AccuracyInput<B>> for SequenceOutput<B> {
    fn adapt(&self) -> AccuracyInput<B> {
        AccuracyInput::new(self.flat_logits(), self.flat_targets())
    }
}

impl<B: Backend> Adaptor<TopKAccuracyInput<B>> for SequenceOutput<B> {
    fn adapt(&self) -> TopKAccuracyInput<B> {
        TopKAccuracyInput::new(self.flat_logits(), self.flat_targets())
    }
}

impl<B: Backend> Adaptor<PerplexityInput<B>> for SequenceOutput<B> {
    fn adapt(&self) -> PerplexityInput<B> {
        PerplexityInput::new(self.flat_logits(), self.flat_targets())
    }
}