File size: 9,322 Bytes
6e8a458 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 | """Generate a cross-model leaderboard from benchmark results.
Reads all result directories, computes scores, and outputs a Markdown
leaderboard table grouped by exam+year.
Usage:
uv run python scripts/generate_leaderboard.py
uv run python scripts/generate_leaderboard.py --results-dir results --min-questions 10
uv run python scripts/generate_leaderboard.py --output leaderboard.json
"""
import argparse
import json
import os
import re
import sys
from collections import defaultdict
# Known exam names that may contain underscores
KNOWN_EXAMS = {"JEE_ADVANCED", "JEE_MAIN", "NEET"}
def parse_result_dirname(dirname: str) -> dict | None:
"""Parse a result directory name into model, exam, year, timestamp.
Format: {model_id}_{EXAM_NAME}_{YEAR}_{YYYYMMDD}_{HHMMSS}
where model_id has slashes replaced with underscores.
Returns dict with keys: model, exam, year, timestamp, or None if unparseable.
"""
parts = dirname.split("_")
if len(parts) < 5:
return None
# Last 2 parts are timestamp
timestamp = "_".join(parts[-2:])
# Validate timestamp format
if not (len(parts[-2]) == 8 and parts[-2].isdigit() and
len(parts[-1]) == 6 and parts[-1].isdigit()):
return None
# Year is parts[-3]
year = parts[-3]
if not (len(year) == 4 and year.isdigit()):
return None
# Exam name: check if it's a 2-part exam (JEE_ADVANCED, JEE_MAIN)
if len(parts) >= 6 and parts[-4] in ("ADVANCED", "MAIN"):
exam = parts[-5] + "_" + parts[-4]
model_parts = parts[:-5]
else:
exam = parts[-4]
model_parts = parts[:-4]
if exam not in KNOWN_EXAMS:
return None
if not model_parts:
return None
# Reconstruct model name: first underscore -> slash (provider/model)
model_name = "_".join(model_parts)
first_underscore = model_name.find("_")
if first_underscore > 0:
model_name = model_name[:first_underscore] + "/" + model_name[first_underscore + 1:]
return {
"model": model_name,
"exam": exam,
"year": year,
"timestamp": timestamp,
}
def load_summary_jsonl(filepath: str) -> list[dict]:
"""Load records from a summary.jsonl file."""
records = []
with open(filepath, "r") as f:
for line in f:
line = line.strip()
if line:
records.append(json.loads(line))
return records
def extract_max_score_from_md(filepath: str) -> int | None:
"""Extract max possible score from summary.md's Overall Score line.
Handles both formats:
- New: **Overall Score:** **315** / **360**
- Old: **Overall Score:** **727 / 800**
"""
try:
with open(filepath, "r") as f:
for line in f:
# New format: **315** / **360**
match = re.search(r"\*\*Overall Score:\*\*\s+\*\*\d+\*\*\s*/\s*\*\*(\d+)\*\*", line)
if match:
return int(match.group(1))
# Old format: **727 / 800**
match = re.search(r"\*\*Overall Score:\*\*\s+\*\*\d+\s*/\s*(\d+)\*\*", line)
if match:
return int(match.group(1))
# Oldest format: **322** (Max score is 360)
match = re.search(r"\(Max score is (\d+)\)", line)
if match:
return int(match.group(1))
except (OSError, ValueError):
pass
return None
def compute_stats(records: list[dict]) -> dict:
"""Compute aggregate stats from summary.jsonl records."""
total_score = sum(r.get("marks_awarded", 0) for r in records)
correct = sum(1 for r in records if r.get("evaluation_status") in ("correct", "correct_full"))
partial = sum(1 for r in records if r.get("evaluation_status", "").startswith("partial_"))
incorrect = sum(1 for r in records if r.get("evaluation_status") in ("incorrect", "incorrect_negative"))
skipped = sum(1 for r in records if r.get("evaluation_status") == "skipped")
failures = sum(1 for r in records if r.get("evaluation_status") in (
"failure_api_or_parse", "failure_unexpected_type", "error_bad_ground_truth"))
return {
"score": total_score,
"correct": correct,
"partial": partial,
"incorrect": incorrect,
"skipped": skipped,
"failures": failures,
"num_questions": len(records),
}
def load_summary_json(filepath: str) -> dict | None:
"""Load stats from an old-format summary.json file."""
try:
with open(filepath, "r") as f:
data = json.load(f)
return {
"score": data.get("overall_score", 0),
"correct": data.get("overall_correct", data.get("overall_correct_full", 0)),
"partial": data.get("overall_partial_correct", 0),
"incorrect": data.get("overall_incorrect", data.get("overall_incorrect_choice", 0)),
"skipped": data.get("overall_skipped", 0),
"failures": data.get("overall_api_parse_failures", 0),
"num_questions": data.get("total_questions_processed", 0),
}
except (OSError, json.JSONDecodeError, KeyError):
return None
def scan_results(results_dir: str, min_questions: int) -> list[dict]:
"""Scan result directories and collect stats."""
entries = []
if not os.path.isdir(results_dir):
print(f"Results directory not found: {results_dir}", file=sys.stderr)
return entries
for dirname in sorted(os.listdir(results_dir)):
dirpath = os.path.join(results_dir, dirname)
if not os.path.isdir(dirpath):
continue
parsed = parse_result_dirname(dirname)
if not parsed:
continue
# Try summary.jsonl first (new format), then summary.json (old format)
stats = None
summary_jsonl_path = os.path.join(dirpath, "summary.jsonl")
summary_json_path = os.path.join(dirpath, "summary.json")
if os.path.exists(summary_jsonl_path):
records = load_summary_jsonl(summary_jsonl_path)
if records:
stats = compute_stats(records)
elif os.path.exists(summary_json_path):
stats = load_summary_json(summary_json_path)
if not stats or stats.get("num_questions", 0) < min_questions:
continue
# Try to get max score from summary.md
md_path = os.path.join(dirpath, "summary.md")
max_score = extract_max_score_from_md(md_path)
entry = {
**parsed,
**stats,
"max_score": max_score,
"result_dir": dirname,
}
entries.append(entry)
return entries
def generate_markdown(entries: list[dict]) -> str:
"""Generate a Markdown leaderboard table grouped by exam+year."""
if not entries:
return "# Benchmark Leaderboard\n\nNo results found.\n"
# Group by exam+year
groups: dict[str, list[dict]] = defaultdict(list)
for e in entries:
key = f"{e['exam']}_{e['year']}"
groups[key].append(e)
lines = ["# Benchmark Leaderboard\n"]
for group_key in sorted(groups.keys()):
group_entries = groups[group_key]
# Sort by score descending
group_entries.sort(key=lambda x: x["score"], reverse=True)
exam_display = group_key.replace("_", " ").replace("JEE ADVANCED", "JEE Advanced").replace("JEE MAIN", "JEE Main")
lines.append(f"\n## {exam_display}\n")
lines.append("| Rank | Model | Score | Max | % | Correct | Partial | Incorrect | Skipped | Failures |")
lines.append("|------|-------|-------|-----|---|---------|---------|-----------|---------|----------|")
for rank, e in enumerate(group_entries, 1):
max_score = e.get("max_score") or "?"
if isinstance(max_score, int) and max_score > 0:
pct = f"{e['score'] / max_score * 100:.1f}%"
else:
pct = "?"
lines.append(
f"| {rank} | {e['model']} | {e['score']} | {max_score} | {pct} "
f"| {e['correct']} | {e['partial']} | {e['incorrect']} | {e['skipped']} | {e['failures']} |"
)
lines.append("")
return "\n".join(lines)
def main():
parser = argparse.ArgumentParser(description="Generate benchmark leaderboard.")
parser.add_argument(
"--results-dir",
type=str,
default="results",
help="Path to the results directory (default: results).",
)
parser.add_argument(
"--min-questions",
type=int,
default=10,
help="Minimum questions to include a run (default: 10, filters incomplete runs).",
)
parser.add_argument(
"--output",
type=str,
help="Output path for leaderboard.json.",
)
args = parser.parse_args()
entries = scan_results(args.results_dir, args.min_questions)
if not entries:
print("No valid results found.", file=sys.stderr)
sys.exit(1)
md = generate_markdown(entries)
print(md)
if args.output:
with open(args.output, "w") as f:
json.dump(entries, f, indent=2)
print(f"\nJSON output saved to {args.output}", file=sys.stderr)
if __name__ == "__main__":
main()
|