burn_dragon_language 0.5.0

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
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179

mod factory;
mod huggingface;
mod local_text;
pub mod scheduler;
mod shakespeare;
mod universality;

use crate::tokenizer::SharedTokenizer;

pub use factory::build_dataset;
pub use huggingface::HuggingFaceDataset;
pub use local_text::LocalTextDataset;
pub use scheduler::{
    RandomDataLoader, SequenceBatch, StreamingDataLoader, TokenSequenceDataset,
    sample_batch_with_shape,
};
pub use shakespeare::ShakespeareDataset;
pub use universality::UniversalityDataset;

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

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

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

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

    pub fn from_local_text(dataset: LocalTextDataset) -> Self {
        Self::LocalText(dataset)
    }

    pub fn from_universality(dataset: UniversalityDataset) -> Self {
        Self::Universality(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::LocalText(dataset) => dataset.tokenizer(),
            Dataset::HuggingFace(dataset) => dataset.tokenizer(),
            Dataset::Universality(dataset) => dataset.tokenizer(),
        }
    }

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

    fn copy_token_range(&self, start: usize, dst: &mut [u32]) {
        match self {
            Dataset::Shakespeare(dataset) => dataset.copy_token_range(start, dst),
            Dataset::LocalText(dataset) => dataset.copy_token_range(start, dst),
            Dataset::HuggingFace(dataset) => dataset.copy_token_range(start, dst),
            Dataset::Universality(dataset) => dataset.copy_token_range(start, dst),
        }
    }

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

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

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

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

    fn preferred_logical_document_tokens(&self, split: DatasetSplit) -> Option<usize> {
        match self {
            Dataset::Shakespeare(dataset) => dataset.preferred_logical_document_tokens(split),
            Dataset::LocalText(dataset) => dataset.preferred_logical_document_tokens(split),
            Dataset::HuggingFace(dataset) => dataset.preferred_logical_document_tokens(split),
            Dataset::Universality(dataset) => dataset.preferred_logical_document_tokens(split),
        }
    }

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

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

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