File size: 6,686 Bytes
a2755fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
180
181
import gc

from datasets import load_dataset
from litdata import optimize, TokensLoader
from litgpt.tokenizer import Tokenizer
from functools import partial


def batch_iterator(name=None):
    # code
    if name in (None, 'bigcode/programming-languages-keywords'):
        dataset = load_dataset('bigcode/programming-languages-keywords', split='train')

        for row in dataset:
            for n in row['keywords']:
                yield n

        del dataset
        gc.collect()

    # code
    if name in (None, 'bigcode/the-stack-smol-xs'):
        dataset = (
            load_dataset('bigcode/the-stack-smol-xs', lang, split='train', trust_remote_code=True)
            for lang in [
                'ada', 'agda', 'alloy', 'antlr', 'applescript', 'assembly', 'augeas', 'awk', 'batchfile', 'bison', 'bluespec', 'c',
                'c++', 'c-sharp', 'clojure', 'cmake', 'coffeescript', 'common-lisp', 'css', 'cuda', 'dart', 'dockerfile', 'elixir',
                'elm', 'emacs-lisp','erlang', 'f-sharp', 'fortran', 'glsl', 'go', 'groovy', 'haskell','html', 'idris', 'isabelle', 'java', 
                'java-server-pages', 'javascript', 'julia', 'kotlin', 'lean', 'literate-agda', 'literate-coffeescript', 'literate-haskell',
                'lua', 'makefile', 'maple', 'markdown', 'mathematica', 'matlab', 'ocaml', 'pascal', 'perl', 'php', 'powershell', 'prolog',
                'protocol-buffer', 'python', 'r', 'racket', 'restructuredtext', 'rmarkdown', 'ruby', 'rust', 'sas', 'scala', 'scheme', 
                'shell', 'smalltalk', 'solidity', 'sparql', 'sql', 'stan', 'standard-ml', 'stata', 'systemverilog', 'tcl', 'tcsh', 'tex', 
                'thrift', 'typescript', 'verilog', 'vhdl', 'visual-basic', 'xslt', 'yacc', 'zig'
            ]
        )

        for d in dataset:
            for row in d:
                yield row['content']

        del dataset
        gc.collect()

    # text
    if name in (None, 'nampdn-ai/tiny-textbooks'):
        dataset = load_dataset('nampdn-ai/tiny-textbooks', split='train')

        for row in dataset:
            yield row['text']

        del dataset
        gc.collect()

    # text
    if name in (None, 'xu-song/cc100-samples'):
        dataset = (
            load_dataset('xu-song/cc100-samples', lang, split='train')
            for lang in ['am', 'ar', 'as', 'az', 'be', 'bg', 'bn', 'bn_rom', 'br', 'bs', 'ca', 'cs', 'cy', 'da', 'de', 'el', 'en', 'eo', 'es', 'et', 'eu', 'fa', 'ff', 'fi', 'fr', 'fy', 'ga', 'gd', 'gl', 'gn', 'gu', 'ha', 'he', 'hi', 'hi_rom', 'hr', 'ht', 'hu', 'hy', 'id', 'ig', 'is', 'it', 'ja', 'jv', 'ka', 'kk', 'km', 'kn', 'ko', 'ku', 'ky', 'la', 'lg', 'li', 'ln', 'lo', 'lt', 'lv', 'mg', 'mk', 'ml', 'mn', 'mr', 'ms', 'my', 'my_zaw', 'ne', 'nl', 'no', 'ns', 'om', 'or', 'pa', 'pl', 'ps', 'pt', 'qu', 'rm', 'ro', 'ru', 'sa', 'si', 'sc', 'sd', 'sk', 'sl', 'so', 'sq', 'sr', 'ss', 'su', 'sv', 'sw', 'ta', 'ta_rom', 'te', 'te_rom', 'th', 'tl', 'tn', 'tr', 'ug', 'uk', 'ur', 'ur_rom', 'uz', 'vi', 'wo', 'xh', 'yi', 'yo', 'zh-Hans', 'zh-Hant', 'zu']
        )

        for d in dataset:
            for row in d['text']:
                yield row

        del dataset
        gc.collect()

    # code
    if name in (None, 'm-a-p/CodeFeedback-Filtered-Instruction'):
        dataset = load_dataset('m-a-p/CodeFeedback-Filtered-Instruction', split='train')

        for row in dataset:
            yield row['query'] + '\n' + row['answer']

        del dataset
        gc.collect()

    # code
    if name in (None, 'nampdn-ai/tiny-codes'):
        dataset = load_dataset('nampdn-ai/tiny-codes', split='train')
        
        for row in dataset:
            yield row['prompt'] + '\n' + row['response']
        
        del dataset
        gc.collect()

    # math
    if name in (None, 'ajibawa-2023/Maths-College'):
        dataset = load_dataset('ajibawa-2023/Maths-College', split='train')
        
        for row in dataset:
            yield row['instruction'] + '\n' + row['output']
        
        del dataset
        gc.collect()

    # math
    if name in (None, 'microsoft/orca-math-word-problems-200k'):
        dataset = load_dataset('microsoft/orca-math-word-problems-200k', split='train')

        for row in dataset:
            yield row['question'] + '\n' + row['answer']

        del dataset
        gc.collect()

    # text
    if name in (None, 'mlabonne/FineTome-100k'):
        dataset = load_dataset('mlabonne/FineTome-100k', split='train')

        for row in dataset['conversations']:
            yield '\n'.join(n['value'] for n in row)

        del dataset
        gc.collect()

    # instruction
    if name in (None, 'arcee-ai/agent-data'):
        dataset = load_dataset('arcee-ai/agent-data', split='train')
        
        for row in dataset['conversations']:
            yield '\n'.join(n['value'] for n in row)
        
        del dataset
        gc.collect()

    # instruction
    if name in (None, 'cognitivecomputations/SystemChat-2.0'):
        dataset = (
            load_dataset('cognitivecomputations/SystemChat-2.0', data_files='SystemChat_filtered.jsonl', split='train'),
            load_dataset('cognitivecomputations/SystemChat-2.0', data_files='SystemChat_multilingual.jsonl', split='train'),
        )
        
        for d in dataset:
            for row in d['messages']:
                yield '\n'.join(n['content'] for n in row)
        
        del dataset
        gc.collect()

    # emoji
    if name in (None, 'badrex/llm-emoji-dataset'):
        dataset = load_dataset('badrex/llm-emoji-dataset', split='train')
        
        for row in dataset:
            yield f'{row["character"]}\n{row["unicode"]}\n{row["short description"]}\n{row["tags"]}\n{row["LLM description"]}'
        
        del dataset
        gc.collect()


def tokenize_fn(dataset_name, tokenizer=None):
    for text in batch_iterator(dataset_name):
        text_ids = tokenizer.encode(text, bos=False, eos=True)
        yield text_ids

datasets_names = [
    'bigcode/programming-languages-keywords',
    'bigcode/the-stack-smol-xs',
    'nampdn-ai/tiny-textbooks',
    'xu-song/cc100-samples',
    'm-a-p/CodeFeedback-Filtered-Instruction',
    'nampdn-ai/tiny-codes',
    'ajibawa-2023/Maths-College',
    'microsoft/orca-math-word-problems-200k',
    'mlabonne/FineTome-100k',
    'arcee-ai/agent-data',
    'cognitivecomputations/SystemChat-2.0',
    'badrex/llm-emoji-dataset',
]

outputs = optimize(
    fn=partial(tokenize_fn, tokenizer=Tokenizer('..')),
    inputs=datasets_names,
    output_dir='../data/',
    # Number of tokens to store by chunks. This is roughly 64MB of tokens per chunk.
    chunk_size=(2049 * 8012),
    num_workers=16,
)