Chandan Akiti
commited on
Commit
•
c705738
1
Parent(s):
bf61ae3
Upload RedPajama-Data-1T-no-cc-c4.py
Browse files- RedPajama-Data-1T-no-cc-c4.py +157 -0
RedPajama-Data-1T-no-cc-c4.py
ADDED
@@ -0,0 +1,157 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Copyright 2023 Together Computer
|
2 |
+
#
|
3 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
+
# you may not use this file except in compliance with the License.
|
5 |
+
# You may obtain a copy of the License at
|
6 |
+
#
|
7 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
8 |
+
#
|
9 |
+
# Unless required by applicable law or agreed to in writing, software
|
10 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
11 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
+
# See the License for the specific language governing permissions and
|
13 |
+
# limitations under the License.
|
14 |
+
|
15 |
+
# Lint as: python3
|
16 |
+
"""RedPajama: An Open-Source, Clean-Room 1.2 Trillion Token Dataset."""
|
17 |
+
|
18 |
+
|
19 |
+
import json
|
20 |
+
|
21 |
+
import datasets
|
22 |
+
import traceback
|
23 |
+
|
24 |
+
logger = datasets.logging.get_logger(__name__)
|
25 |
+
|
26 |
+
|
27 |
+
_DESCRIPTION = """\
|
28 |
+
RedPajama is a clean-room, fully open-source implementation of the LLaMa dataset.
|
29 |
+
"""
|
30 |
+
|
31 |
+
_URL_LISTS = {
|
32 |
+
"arxiv": "urls/arxiv.txt",
|
33 |
+
"book": "urls/book.txt",
|
34 |
+
"github": "urls/github.txt",
|
35 |
+
"stackexchange": "urls/stackexchange.txt",
|
36 |
+
"wikipedia": "urls/wikipedia.txt",
|
37 |
+
}
|
38 |
+
|
39 |
+
|
40 |
+
class RedPajama1TConfig(datasets.BuilderConfig):
|
41 |
+
"""BuilderConfig for RedPajama sample."""
|
42 |
+
|
43 |
+
def __init__(self, *args, subsets, **kwargs):
|
44 |
+
"""BuilderConfig for RedPajama.
|
45 |
+
Args:
|
46 |
+
**kwargs: keyword arguments forwarded to super.
|
47 |
+
"""
|
48 |
+
super(RedPajama1TConfig, self).__init__(**kwargs)
|
49 |
+
|
50 |
+
self.subsets = subsets
|
51 |
+
|
52 |
+
|
53 |
+
class RedPajama1T(datasets.GeneratorBasedBuilder):
|
54 |
+
"""RedPajama: Reproducing the LLaMA training dataset of over 1.2 trillion tokens. Version 1.0.0."""
|
55 |
+
|
56 |
+
BUILDER_CONFIGS = [
|
57 |
+
RedPajama1TConfig(
|
58 |
+
subsets = list(_URL_LISTS.keys()),
|
59 |
+
name="plain_text",
|
60 |
+
version=datasets.Version("1.0.0", ""),
|
61 |
+
description="Plain text",
|
62 |
+
),
|
63 |
+
]
|
64 |
+
|
65 |
+
def _info(self):
|
66 |
+
return datasets.DatasetInfo(
|
67 |
+
description=_DESCRIPTION,
|
68 |
+
features=datasets.Features(
|
69 |
+
{
|
70 |
+
"text": datasets.Value("string"),
|
71 |
+
"meta": datasets.Value("string"),
|
72 |
+
"red_pajama_subset": datasets.Value("string"),
|
73 |
+
}
|
74 |
+
),
|
75 |
+
supervised_keys=None,
|
76 |
+
)
|
77 |
+
|
78 |
+
def _split_generators(self, dl_manager):
|
79 |
+
url_lists = dl_manager.download_and_extract({
|
80 |
+
subset: _URL_LISTS[subset] for subset in self.config.subsets
|
81 |
+
})
|
82 |
+
|
83 |
+
urls = {}
|
84 |
+
|
85 |
+
for subset, url_list in url_lists.items():
|
86 |
+
with open(url_list, encoding="utf-8") as f:
|
87 |
+
urls[subset] = [line.strip() for line in f][:1]
|
88 |
+
|
89 |
+
downloaded_files = dl_manager.download(urls)
|
90 |
+
|
91 |
+
return [
|
92 |
+
datasets.SplitGenerator(
|
93 |
+
name=datasets.Split.TRAIN,
|
94 |
+
gen_kwargs = {
|
95 |
+
"files": {
|
96 |
+
subset: downloaded_files[subset]
|
97 |
+
for subset in self.config.subsets
|
98 |
+
}
|
99 |
+
}
|
100 |
+
)
|
101 |
+
]
|
102 |
+
|
103 |
+
def _generate_examples(self, files):
|
104 |
+
"""This function returns the examples in the raw (text) form."""
|
105 |
+
key = 0
|
106 |
+
for subset in files:
|
107 |
+
if subset == "common_crawl":
|
108 |
+
import zstandard as zstd
|
109 |
+
|
110 |
+
for path in files[subset]:
|
111 |
+
with zstd.open(open(path, "rb"), "rt", encoding="utf-8") as f:
|
112 |
+
for i, row in enumerate(f):
|
113 |
+
try:
|
114 |
+
data = json.loads(row)
|
115 |
+
text = data["text"]
|
116 |
+
del data["text"]
|
117 |
+
yield key, {
|
118 |
+
"text": text,
|
119 |
+
"meta": json.dumps(data),
|
120 |
+
"red_pajama_subset": subset,
|
121 |
+
}
|
122 |
+
key += 1
|
123 |
+
except Exception as e:
|
124 |
+
print(f'Subset: {subset}')
|
125 |
+
print(f'Path: {path}')
|
126 |
+
print(f'Row: {row}')
|
127 |
+
traceback.print_exc()
|
128 |
+
|
129 |
+
raise e
|
130 |
+
else:
|
131 |
+
for path in files[subset]:
|
132 |
+
with open(path, encoding="utf-8") as f:
|
133 |
+
for i, row in enumerate(f):
|
134 |
+
try:
|
135 |
+
data = json.loads(row)
|
136 |
+
if "meta" not in data:
|
137 |
+
text = data["text"]
|
138 |
+
del data["text"]
|
139 |
+
yield key, {
|
140 |
+
"text": text,
|
141 |
+
"meta": json.dumps(data),
|
142 |
+
"red_pajama_subset": subset,
|
143 |
+
}
|
144 |
+
else:
|
145 |
+
yield key, {
|
146 |
+
"text": data["text"],
|
147 |
+
"meta": data["meta"],
|
148 |
+
"red_pajama_subset": subset,
|
149 |
+
}
|
150 |
+
key += 1
|
151 |
+
except Exception as e:
|
152 |
+
print(f'Subset: {subset}')
|
153 |
+
print(f'Path: {path}')
|
154 |
+
print(f'Row: {row}')
|
155 |
+
traceback.print_exc()
|
156 |
+
|
157 |
+
raise e
|