File size: 7,174 Bytes
4f040da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52e8aa2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4f040da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52e8aa2
 
 
 
 
4f040da
 
 
 
 
52e8aa2
4f040da
52e8aa2
 
4f040da
 
 
 
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
#!/usr/bin/env python3
"""Audit pinned task datasets without assigning train/evaluation roles."""

from __future__ import annotations

import argparse
import csv
import hashlib
import json
import re
from pathlib import Path

import pyarrow.parquet as pq


def sha256(path: Path) -> str:
    digest = hashlib.sha256()
    with path.open("rb") as handle:
        for chunk in iter(lambda: handle.read(1024 * 1024), b""):
            digest.update(chunk)
    return digest.hexdigest()


def write_csv(path: Path, rows: list[dict[str, object]]) -> None:
    path.parent.mkdir(parents=True, exist_ok=True)
    with path.open("w", newline="") as handle:
        writer = csv.DictWriter(handle, fieldnames=list(rows[0]))
        writer.writeheader()
        writer.writerows(rows)


def audit_swe(path: Path, output: Path, revision: str) -> None:
    table = pq.read_table(path, columns=["language", "repo", "base_commit", "license"])
    data = table.to_pydict()
    languages = sorted(set(data["language"]))
    rows = []
    for language in languages:
        indices = [i for i, value in enumerate(data["language"]) if value == language]
        rows.append(
            {
                "language": language,
                "tasks": len(indices),
                "unique_repositories": len({data["repo"][i] for i in indices}),
                "unique_base_commits": len({data["base_commit"][i] for i in indices}),
                "license_values": len({data["license"][i] for i in indices}),
                "source_revision": revision,
                "source_sha256": sha256(path),
            }
        )
    rows.sort(key=lambda row: (-int(row["tasks"]), str(row["language"])))
    write_csv(output, rows)


def audit_mceval_instruct(path: Path, output: Path, revision: str) -> None:
    records = json.loads(path.read_text())
    grouped: dict[str, list[dict[str, str]]] = {}
    for record in records:
        grouped.setdefault(record["language"].casefold(), []).append(record)
    rows = []
    for language, group in grouped.items():
        rows.append(
            {
                "normalized_language": language,
                "rows": len(group),
                "unique_instructions": len({record["instruction"] for record in group}),
                "unique_outputs": len({record["output"] for record in group}),
                "output_utf8_bytes": sum(len(record["output"].encode()) for record in group),
                "original_labels": ";".join(sorted({record["language"] for record in group})),
                "source_revision": revision,
                "source_sha256": sha256(path),
            }
        )
    rows.sort(key=lambda row: (-int(row["rows"]), str(row["normalized_language"])))
    write_csv(output, rows)


def audit_mceval_eval(root: Path, output: Path, revision: str) -> None:
    grouped: dict[tuple[str, str], dict[str, object]] = {}
    source_files = sorted(root.glob("generation/*.jsonl"))
    source_files += sorted(root.glob("explanation/*.jsonl"))
    source_files += sorted(root.glob("completion/*/*.jsonl"))
    for path in source_files:
        relative = path.relative_to(root)
        surface = "/".join(relative.parts[:-1])
        language = path.stem
        group = grouped.setdefault(
            (surface, language),
            {"rows": 0, "base_ids": set(), "solutions": set(), "tests": set()},
        )
        with path.open() as handle:
            for line in handle:
                record = json.loads(line)
                group["rows"] = int(group["rows"]) + 1
                match = re.match(r"^([^/]+)/(\d+)", record["task_id"])
                group["base_ids"].add(match.group(2) if match else record["task_id"])
                group["solutions"].add(record.get("canonical_solution", ""))
                group["tests"].add(record.get("test", ""))
    tree_hash = hashlib.sha256()
    for path in source_files:
        tree_hash.update(str(path.relative_to(root)).encode())
        tree_hash.update(bytes.fromhex(sha256(path)))
    rows = []
    for (surface, language), group in grouped.items():
        rows.append(
            {
                "surface": surface,
                "language": language,
                "rows": group["rows"],
                "unique_base_problem_ids": len(group["base_ids"]),
                "unique_canonical_solutions": len(group["solutions"]),
                "unique_tests": len(group["tests"]),
                "source_revision": revision,
                "audited_tree_sha256": tree_hash.hexdigest(),
            }
        )
    rows.sort(key=lambda row: (str(row["surface"]), str(row["language"])))
    write_csv(output, rows)


def audit_swe_leaderboard(root: Path, output: Path, revision: str) -> None:
    """Inventory every frozen leaderboard split without treating it as training data."""
    source_files = sorted((root / "data").glob("*.parquet"))
    if not source_files:
        raise FileNotFoundError(f"no parquet files found under {root / 'data'}")
    rows = []
    for path in source_files:
        table = pq.read_table(path, columns=["repo", "instance_id", "base_commit"])
        data = table.to_pydict()
        rows.append(
            {
                "split": path.stem.split("-00000", 1)[0],
                "rows": table.num_rows,
                "unique_instances": len(set(data["instance_id"])),
                "unique_repositories": len(set(data["repo"])),
                "unique_base_commits": len(set(data["base_commit"])),
                "source_revision": revision,
                "file_sha256": sha256(path),
            }
        )
    write_csv(output, rows)


def main() -> None:
    parser = argparse.ArgumentParser()
    subparsers = parser.add_subparsers(dest="source", required=True)

    swe = subparsers.add_parser("swe-rebench-v2")
    swe.add_argument("--input", type=Path, required=True)
    swe.add_argument("--output", type=Path, required=True)
    swe.add_argument("--revision", required=True)

    mceval = subparsers.add_parser("mceval-instruct")
    mceval.add_argument("--input", type=Path, required=True)
    mceval.add_argument("--output", type=Path, required=True)
    mceval.add_argument("--revision", required=True)

    mceval_eval = subparsers.add_parser("mceval-eval")
    mceval_eval.add_argument("--input", type=Path, required=True)
    mceval_eval.add_argument("--output", type=Path, required=True)
    mceval_eval.add_argument("--revision", required=True)

    leaderboard = subparsers.add_parser("swe-rebench-leaderboard")
    leaderboard.add_argument("--input", type=Path, required=True)
    leaderboard.add_argument("--output", type=Path, required=True)
    leaderboard.add_argument("--revision", required=True)

    args = parser.parse_args()
    if args.source == "swe-rebench-v2":
        audit_swe(args.input, args.output, args.revision)
    elif args.source == "mceval-instruct":
        audit_mceval_instruct(args.input, args.output, args.revision)
    elif args.source == "mceval-eval":
        audit_mceval_eval(args.input, args.output, args.revision)
    else:
        audit_swe_leaderboard(args.input, args.output, args.revision)


if __name__ == "__main__":
    main()