burn_dragon_language 0.4.0

Language modeling components 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

mod factory;
mod huggingface;
pub mod scheduler;
mod shakespeare;

use crate::tokenizer::SharedTokenizer;

pub use factory::build_dataset;
pub use huggingface::HuggingFaceDataset;
pub use scheduler::{RandomDataLoader, SequenceBatch, TokenSequenceDataset};
pub use shakespeare::ShakespeareDataset;

#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum DatasetSplit {
    Train,
    Val,
}

#[derive(Clone)]
pub enum Dataset {
    Shakespeare(ShakespeareDataset),
    HuggingFace(HuggingFaceDataset),
}

impl Dataset {
    pub fn from_shakespeare(dataset: ShakespeareDataset) -> Self {
        Self::Shakespeare(dataset)
    }

    pub fn from_huggingface(dataset: HuggingFaceDataset) -> Self {
        Self::HuggingFace(dataset)
    }

    pub fn tokenizer(&self) -> SharedTokenizer {
        TokenSequenceDataset::tokenizer(self)
    }

    pub fn train_split_ratio(&self) -> f32 {
        TokenSequenceDataset::train_split_ratio(self)
    }

    pub fn batch_size(&self) -> usize {
        TokenSequenceDataset::batch_size(self)
    }

    pub fn steps_per_epoch(&self, split: DatasetSplit) -> usize {
        TokenSequenceDataset::steps_per_epoch(self, split)
    }
}

impl TokenSequenceDataset for Dataset {
    fn tokenizer(&self) -> SharedTokenizer {
        match self {
            Dataset::Shakespeare(dataset) => dataset.tokenizer(),
            Dataset::HuggingFace(dataset) => dataset.tokenizer(),
        }
    }

    fn tokens(&self) -> &[u32] {
        match self {
            Dataset::Shakespeare(dataset) => dataset.tokens(),
            Dataset::HuggingFace(dataset) => dataset.tokens(),
        }
    }

    fn train_len(&self) -> usize {
        match self {
            Dataset::Shakespeare(dataset) => dataset.train_len(),
            Dataset::HuggingFace(dataset) => dataset.train_len(),
        }
    }

    fn block_size(&self) -> usize {
        match self {
            Dataset::Shakespeare(dataset) => dataset.block_size(),
            Dataset::HuggingFace(dataset) => dataset.block_size(),
        }
    }

    fn batch_size(&self) -> usize {
        match self {
            Dataset::Shakespeare(dataset) => dataset.batch_size(),
            Dataset::HuggingFace(dataset) => dataset.batch_size(),
        }
    }

    fn train_split_ratio(&self) -> f32 {
        match self {
            Dataset::Shakespeare(dataset) => dataset.train_split_ratio(),
            Dataset::HuggingFace(dataset) => dataset.train_split_ratio(),
        }
    }

    fn split_offset_and_span(&self, split: DatasetSplit) -> (usize, usize) {
        match self {
            Dataset::Shakespeare(dataset) => {
                TokenSequenceDataset::split_offset_and_span(dataset, split)
            }
            Dataset::HuggingFace(dataset) => {
                TokenSequenceDataset::split_offset_and_span(dataset, split)
            }
        }
    }

    fn steps_per_epoch(&self, split: DatasetSplit) -> usize {
        match self {
            Dataset::Shakespeare(dataset) => TokenSequenceDataset::steps_per_epoch(dataset, split),
            Dataset::HuggingFace(dataset) => TokenSequenceDataset::steps_per_epoch(dataset, split),
        }
    }

    fn decode(&self, tokens: &[i64]) -> String {
        match self {
            Dataset::Shakespeare(dataset) => TokenSequenceDataset::decode(dataset, tokens),
            Dataset::HuggingFace(dataset) => TokenSequenceDataset::decode(dataset, tokens),
        }
    }
}

pub type ShakespeareSplit = DatasetSplit;
pub type ShakespeareBatch<B> = SequenceBatch<B>;
pub type ShakespeareRandomDataLoader<B> = RandomDataLoader<B>;