AtlasUnified commited on
Commit
4649fcf
·
1 Parent(s): 841e5fd

Create atlas-preprocessed-code.py

Browse files
Files changed (1) hide show
  1. atlas-preprocessed-code.py +125 -0
atlas-preprocessed-code.py ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import json
2
+ import datasets
3
+ import traceback
4
+ import os
5
+
6
+ logger = datasets.logging.get_logger(__name__)
7
+
8
+ _DESCRIPTION = """\
9
+ RedPajama is a clean-room, fully open-source implementation of the LLaMa dataset.
10
+ """
11
+
12
+ # Define a list of programming languages for the subsets
13
+ _LANGUAGE_SUBSETS = ["csharp", "cplusplus", "c", "go", "html", "haskell",
14
+ "java", "javascript", "jupyter", "kotlin", "php", "perl",
15
+ "python", "ruby", "rust", "shell", "swift", "typescript", "v"]
16
+
17
+ # Generate URLs for the subsets
18
+ _URL_LISTS = {language: f"https://huggingface.co/datasets/AtlasUnified/atlas-preprocessed-code/blob/main/{language}.jsonl"
19
+ for language in _LANGUAGE_SUBSETS}
20
+
21
+ _URL_BASE = 'https://data.together.xyz/redpajama-data-1T/v1.0.0'
22
+
23
+ _DATA_DIR = os.environ.get('RED_PAJAMA_DATA_DIR', None)
24
+
25
+ class RedPajama1TConfig(datasets.BuilderConfig):
26
+ """BuilderConfig for RedPajama sample."""
27
+
28
+ def __init__(self, *args, subsets, **kwargs):
29
+ """BuilderConfig for RedPajama.
30
+ Args:
31
+ **kwargs: keyword arguments forwarded to super.
32
+ """
33
+ super(RedPajama1TConfig, self).__init__(**kwargs)
34
+ self.subsets = subsets
35
+
36
+ class RedPajama1T(datasets.GeneratorBasedBuilder):
37
+ """RedPajama: Reproducing the LLaMA training dataset of over 1.2 trillion tokens. Version 1.0.0."""
38
+
39
+ # Generate builder configs for each language subset
40
+ BUILDER_CONFIGS = [RedPajama1TConfig(name=language,
41
+ subsets=[language],
42
+ version=datasets.Version("1.0.0", ""),
43
+ description=f"RedPajama1T {language} subset")
44
+ for language in _LANGUAGE_SUBSETS]
45
+
46
+ def _info(self):
47
+ return datasets.DatasetInfo(
48
+ description=_DESCRIPTION,
49
+ features=datasets.Features(
50
+ {
51
+ "text": datasets.Value("string"),
52
+ "meta": datasets.Value("string"),
53
+ "red_pajama_subset": datasets.Value("string"),
54
+ }
55
+ ),
56
+ supervised_keys=None,
57
+ )
58
+
59
+ def _split_generators(self, dl_manager):
60
+ url_lists = dl_manager.download_and_extract({
61
+ subset: _URL_LISTS[subset] for subset in self.config.subsets
62
+ })
63
+
64
+ urls = {}
65
+
66
+ for subset, url_list in url_lists.items():
67
+ with open(url_list, encoding="utf-8") as f:
68
+ urls[subset] = [line.strip() for line in f]
69
+
70
+ if _DATA_DIR is not None:
71
+ print(f'Reading data from {_DATA_DIR}')
72
+ url_prefix_slashes = len(_URL_BASE.split('/'))
73
+ downloaded_files = {
74
+ subset: [
75
+ os.path.join(_DATA_DIR, *url.split('/')[url_prefix_slashes:])
76
+ for url in url_list
77
+ ]
78
+ for subset, url_list in urls.items()
79
+ }
80
+ else:
81
+ downloaded_files = dl_manager.download(urls)
82
+
83
+ return [
84
+ datasets.SplitGenerator(
85
+ name=datasets.Split.TRAIN,
86
+ gen_kwargs = {
87
+ "files": {
88
+ subset: downloaded_files[subset]
89
+ for subset in self.config.subsets
90
+ }
91
+ }
92
+ )
93
+ ]
94
+
95
+ def _generate_examples(self, files):
96
+ """This function returns the examples in the raw (text) form."""
97
+ key = 0
98
+ for subset in files:
99
+ for path in files[subset]:
100
+ with open(path, encoding="utf-8") as f:
101
+ for i, row in enumerate(f):
102
+ try:
103
+ data = json.loads(row)
104
+ if "meta" not in data:
105
+ text = data["text"]
106
+ del data["text"]
107
+ yield key, {
108
+ "text": text,
109
+ "meta": json.dumps(data),
110
+ "red_pajama_subset": subset,
111
+ }
112
+ else:
113
+ yield key, {
114
+ "text": data["text"],
115
+ "meta": data["meta"],
116
+ "red_pajama_subset": subset,
117
+ }
118
+ key += 1
119
+ except Exception as e:
120
+ print(f'Subset: {subset}')
121
+ print(f'Path: {path}')
122
+ print(f'Row: {row}')
123
+ traceback.print_exc()
124
+
125
+ raise e