demo_generate_30.py

"""Generate and print a 30-passage Search-R1 corpus demo."""

import json
import os
import sys
from collections import Counter

import build_search_r1_corpus as corpus_builder


DEMO_PASSAGE_LIMIT = 30
DEMO_OUTPUT_PATH = corpus_builder.BASE_DIR / "search_r1_corpus_demo.jsonl"
DEMO_TEMP_PATH = corpus_builder.BASE_DIR / "search_r1_corpus_demo.jsonl.tmp"


def main() -> None:
    if hasattr(sys.stdout, "reconfigure"):
        sys.stdout.reconfigure(encoding="utf-8", errors="replace")

    corpus_builder.verify_input_files(corpus_builder.INPUT_FILES)

    jsonl_lines = []
    scanned_articles = 0
    retained_articles = 0
    filter_counts = Counter()

    try:
        with DEMO_TEMP_PATH.open(
            "w", encoding="utf-8", newline="\n"
        ) as output:
            for input_path in corpus_builder.INPUT_FILES:
                documents = corpus_builder.load_documents(input_path)

                for article in documents:
                    scanned_articles += 1
                    reason = corpus_builder.initial_filter_reason(article)

                    if reason is not None:
                        filter_counts[reason] += 1
                        continue

                    raw_content = str(article["content"])
                    raw_title = (
                        ""
                        if article.get("title") is None
                        else str(article["title"])
                    )
                    raw_link = (
                        ""
                        if article.get("link") is None
                        else str(article["link"])
                    )

                    cleaned_content = corpus_builder.clean_content(raw_content)
                    passages = corpus_builder.make_passages(cleaned_content)
                    if not passages:
                        filter_counts["cleaned_content_under_50_words"] += 1
                        continue

                    retained_articles += 1
                    title = corpus_builder.clean_title(raw_title, raw_link)

                    for passage in passages:
                        passage_id = str(len(jsonl_lines))
                        record = {
                            "id": passage_id,
                            "contents": f'"{title}"\n{passage}',
                        }
                        line = json.dumps(record, ensure_ascii=False)
                        output.write(line)
                        output.write("\n")
                        jsonl_lines.append(line)

                        if len(jsonl_lines) == DEMO_PASSAGE_LIMIT:
                            break

                    if len(jsonl_lines) == DEMO_PASSAGE_LIMIT:
                        break

                if len(jsonl_lines) == DEMO_PASSAGE_LIMIT:
                    break

        os.replace(DEMO_TEMP_PATH, DEMO_OUTPUT_PATH)
    except BaseException:
        try:
            DEMO_TEMP_PATH.unlink(missing_ok=True)
        except OSError:
            pass
        raise

    print("=" * 80)
    print(f"Search-R1 corpus demo: {len(jsonl_lines)} JSONL records")
    print("=" * 80)
    for line in jsonl_lines:
        print(line)

    print("=" * 80)
    print("Demo summary")
    print("=" * 80)
    print(f"Scanned source articles: {scanned_articles}")
    print(f"Articles contributing passages: {retained_articles}")
    print(f"Filtered while scanning: {sum(filter_counts.values())}")
    for reason, count in sorted(filter_counts.items()):
        print(f"  {reason}: {count}")
    print(f"Generated passages: {len(jsonl_lines)}")
    print(f"Output file: {DEMO_OUTPUT_PATH}")


if __name__ == "__main__":
    main()