ryanse commited on
Commit
88c84e3
·
verified ·
1 Parent(s): 5ee6145

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +177 -0
README.md CHANGED
@@ -48,3 +48,180 @@ configs:
48
  - split: test
49
  path: problem_files/test-*
50
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
48
  - split: test
49
  path: problem_files/test-*
50
  ---
51
+
52
+ # SWE-Bench Verified
53
+
54
+
55
+
56
+ ```python
57
+ import argparse
58
+ from dataclasses import dataclass, asdict
59
+ import datasets
60
+ from pathlib import Path
61
+ import subprocess
62
+ from typing import Dict, List
63
+
64
+ import tqdm
65
+
66
+ from datasets import Dataset
67
+
68
+ import hashlib
69
+
70
+ from dataclasses import dataclass
71
+
72
+
73
+ @dataclass
74
+ class CodebaseFile:
75
+ path: str
76
+ content: str
77
+
78
+
79
+ class SWEBenchProblem:
80
+ def __init__(self, row):
81
+ self._row = row
82
+
83
+ @property
84
+ def repo(self) -> str:
85
+ return self._row["repo"]
86
+
87
+ @property
88
+ def base_commit(self) -> str:
89
+ return self._row["base_commit"]
90
+
91
+
92
+ @property
93
+ def instance_id(self) -> str:
94
+ return self._row["instance_id"]
95
+
96
+ VALID_EXTENSIONS = {"py"}
97
+
98
+
99
+ def hash_file_content(file_content: str) -> str:
100
+ return hashlib.sha256(file_content.encode()).hexdigest()
101
+
102
+
103
+ @dataclass
104
+ class FileInCodebase:
105
+ file_path: str
106
+ content_hash: str
107
+
108
+
109
+ @dataclass
110
+ class CodebaseContent:
111
+ instance_id: str
112
+ files: List[FileInCodebase]
113
+
114
+
115
+ def clone_repos(problems: list[SWEBenchProblem], repos_dir: Path):
116
+ repos_dir.mkdir(exist_ok=False, parents=True)
117
+
118
+ if len(list(repos_dir.iterdir())):
119
+ raise ValueError("Repos dir should be empty")
120
+
121
+ repos = {problem.repo for problem in problems}
122
+ for repo in tqdm.tqdm(repos, desc="Cloning repos"):
123
+ output = subprocess.run(
124
+ ["git", "clone", f"https://github.com/{repo}.git"],
125
+ cwd=repos_dir,
126
+ capture_output=True,
127
+ )
128
+ assert output.returncode == 0
129
+
130
+
131
+ def get_codebase_content(
132
+ problem: SWEBenchProblem, repos_dir: Path, hash_to_content: Dict[str, str]
133
+ ) -> CodebaseContent:
134
+ repo = problem.repo.split("/")[-1]
135
+ repo_path = repos_dir / repo
136
+
137
+ subprocess.run(
138
+ ["git", "checkout", problem.base_commit], cwd=repo_path, capture_output=True
139
+ )
140
+
141
+ contexts = []
142
+
143
+ for file_path in repo_path.rglob("*"):
144
+ if not file_path.is_file:
145
+ continue
146
+
147
+ if file_path.suffix[1:] not in VALID_EXTENSIONS: # [1:] excludes the '.'
148
+ continue
149
+
150
+ try:
151
+ content = file_path.read_text()
152
+ except UnicodeDecodeError:
153
+ # Ignore these files.
154
+ continue
155
+
156
+ content_hash = hash_file_content(content)
157
+ if content_hash not in hash_to_content:
158
+ hash_to_content[content_hash] = content
159
+
160
+ contexts.append(
161
+ FileInCodebase(
162
+ file_path=str(file_path.relative_to(repo_path)),
163
+ content_hash=content_hash,
164
+ )
165
+ )
166
+
167
+ return CodebaseContent(instance_id=problem.instance_id, files=contexts)
168
+
169
+
170
+ @dataclass
171
+ class ContentDatasetElement:
172
+ hash: str
173
+ content: str
174
+
175
+
176
+ def main():
177
+ parser = argparse.ArgumentParser()
178
+ parser.add_argument(
179
+ "--repo_directory",
180
+ type=Path,
181
+ default=Path("/scr/ryanehrlich/swebench_verified_repos"),
182
+ )
183
+ parser.add_argument(
184
+ "--output_dataset_name",
185
+ type=str,
186
+ default="ScalingIntelligence/swe-bench-verified-codebase-content-staging",
187
+ )
188
+
189
+ args = parser.parse_args()
190
+
191
+ dataset = datasets.load_dataset("princeton-nlp/SWE-bench_Verified", split="test")
192
+ problems = [SWEBenchProblem(row) for row in dataset]
193
+
194
+ clone_repos(problems, args.repo_directory)
195
+ hash_to_content = {}
196
+ codebase_content_per_problem = [
197
+ get_codebase_content(problem, args.repo_directory, hash_to_content)
198
+ for problem in tqdm.tqdm(problems, desc="Fetching codebase content")
199
+ ]
200
+
201
+ hash_to_content_in_hf_form = [
202
+ {
203
+ "hash": hash_,
204
+ "content": content,
205
+ }
206
+ for (hash_, content) in hash_to_content.items()
207
+ ]
208
+
209
+ codebase_content_in_hf_form = [
210
+ asdict(problem) for problem in codebase_content_per_problem
211
+ ]
212
+
213
+ file_content_dataset = Dataset.from_list(hash_to_content_in_hf_form, split="test")
214
+ problems_dataset = Dataset.from_list(codebase_content_in_hf_form, split="test")
215
+
216
+ file_content_dataset.push_to_hub(
217
+ args.output_dataset_name, "file_content", private=True, max_shard_size="256MB"
218
+ )
219
+ problems_dataset.push_to_hub(
220
+ args.output_dataset_name, "problem_files", private=True, max_shard_size="256MB"
221
+ )
222
+
223
+
224
+ if __name__ == "__main__":
225
+ main()
226
+ ```
227
+