Python
#!/usr/bin/env python3
"""
Generate a JSONL file from thinkmorph/vilasr directory structure.
"""
import json
import re
from pathlib import Path
def generate_jsonl(results_dir: Path, output_jsonl: Path):
"""Generate JSONL from text_data.json files in subdirectories."""
results = []
# Find all subdirectories with pattern sample_*_sim_*_initial
for subdir in sorted(results_dir.glob("sample_*_sim_*_initial*")):
if not subdir.is_dir():
continue
# Extract sim number from directory name
match = re.search(r"sim_(\d+)_initial", subdir.name, re.I)
if not match:
continue
sim_num = match.group(1)
image_path = f"sim_{sim_num}_initial.png"
# Read text_data.json
text_data_path = subdir / "text_data.json"
if not text_data_path.exists():
continue
try:
j = json.loads(text_data_path.read_text(encoding="utf-8"))
except Exception:
continue
# Extract raw text and answer
raw_text = ""
answer = None
# For thinkmorph: text_outputs is a list
if "text_outputs" in j and isinstance(j["text_outputs"], list):
raw_text = "\n".join(str(t) for t in j["text_outputs"])
# For vilasr: response is a string
elif "response" in j:
raw_text = str(j.get("response", ""))
# Extract <answer>X</answer>
if raw_text:
answer_match = re.search(r"<answer>(\d+)</answer>", raw_text, re.I)
if answer_match:
answer = int(answer_match.group(1))
# Create result entry
result = {
"image_path": image_path,
"model_output": raw_text,
"answer": answer
}
results.append(result)
# Write JSONL
with output_jsonl.open("w", encoding="utf-8") as f:
for result in results:
f.write(json.dumps(result) + "\n")
print(f"Generated {len(results)} entries in {output_jsonl}")
if __name__ == "__main__":
import sys
if len(sys.argv) != 3:
print(f"Usage: {sys.argv[0]} <results_dir> <output_jsonl>")
sys.exit(1)
results_dir = Path(sys.argv[1])
output_jsonl = Path(sys.argv[2])
generate_jsonl(results_dir, output_jsonl)