| |
| """Reconstruct frozen GuoFeng cases from user-supplied official V1 Train files.""" |
|
|
| import argparse |
| import hashlib |
| import json |
| from pathlib import Path |
|
|
|
|
| def digest(path): |
| h = hashlib.sha256() |
| with path.open('rb') as stream: |
| for block in iter(lambda: stream.read(1024 * 1024), b''): |
| h.update(block) |
| return h.hexdigest() |
|
|
|
|
| def read_rows(path): |
| with path.open(encoding='utf-8') as stream: |
| return [json.loads(line) for line in stream if line.strip()] |
|
|
|
|
| def write_rows(path, rows): |
| with path.open('w', encoding='utf-8') as stream: |
| for row in rows: |
| stream.write(json.dumps(row, ensure_ascii=False) + '\n') |
|
|
|
|
| def write_checksums(root): |
| files = sorted(p for p in root.rglob('*') if p.is_file() and p.name != 'SHA256SUMS') |
| (root / 'SHA256SUMS').write_text(''.join( |
| f'{digest(p)} {p.relative_to(root).as_posix()}\n' for p in files |
| ), encoding='utf-8') |
|
|
|
|
| def reconstruct(recipe, source, reference, output): |
| manifest = json.loads((recipe / 'manifest.json').read_text(encoding='utf-8')) |
| for side, path in [('source', source), ('reference', reference)]: |
| if digest(path) != manifest['upstream'][side + '_sha256']: |
| raise ValueError(f'Official {side} file hash differs from the frozen input') |
| if output.exists(): |
| raise FileExistsError(output) |
| unit_specs = read_rows(recipe / 'units.jsonl') |
| by_row = {r['corpus_row']: r for r in unit_specs} |
| values = {} |
| for side, path in [('source', source), ('reference', reference)]: |
| with path.open(encoding='utf-8') as stream: |
| for line_number, line in enumerate(stream, 1): |
| if line_number not in by_row: |
| continue |
| raw = line.rstrip('\r\n') |
| spans = by_row[line_number][side + '_keep_spans'] |
| if any(a < 0 or b < a or b > len(raw) for a, b in spans): |
| raise ValueError(f'Invalid character span at row {line_number}') |
| values.setdefault(line_number, {})[side] = ''.join(raw[a:b] for a, b in spans) |
| if values.keys() != by_row.keys(): |
| raise ValueError('Missing official input rows') |
| by_work = {} |
| for spec in unit_specs: |
| work = spec['document_id'].split(':', 1)[0] |
| by_work.setdefault(work, {})[spec['canonical_alignment_unit_index']] = spec |
| cases, requests, references, alignments = [], [], [], [] |
| prompt = (recipe / 'prompt.txt').read_text(encoding='utf-8') |
| for case in read_rows(recipe / 'case-index.jsonl'): |
| m = case['metadata'] |
| selected = [by_work[case['work_id']][i] for i in range( |
| m['start_alignment_unit'], m['end_alignment_unit_exclusive'])] |
| for side in ['source', 'reference']: |
| case[side] = '\n'.join(values[s['corpus_row']][side] for s in selected) |
| actual = hashlib.sha256(case[side].encode()).hexdigest() |
| if actual != case[side + '_sha256']: |
| raise ValueError(f'Frozen text mismatch: {case["case_id"]} {side}') |
| cases.append(case) |
| requests.append({'case_id': case['case_id'], |
| 'messages': [{'role': 'user', 'content': prompt + '\n\n' + case['source']}], |
| 'metadata': {k: case[k] for k in [ |
| 'track_id', 'split', 'work_id', 'length_band', 'source_sha256']}}) |
| references.append({k: case[k] for k in ['case_id', 'reference', 'reference_sha256']}) |
| for index, spec in enumerate(selected): |
| row = {k: v for k, v in spec.items() if not k.endswith('_keep_spans')} |
| row.update(case_id=case['case_id'], alignment_unit_index=index, |
| **values[spec['corpus_row']]) |
| alignments.append(row) |
| output.mkdir(parents=True) |
| (output / 'segale').mkdir() |
| for name, rows in [('cases.jsonl', cases), ('requests.jsonl', requests), |
| ('references.jsonl', references), ('segale/alignment-units.jsonl', alignments)]: |
| write_rows(output / name, rows) |
| (output / 'segale/document-boundaries.jsonl').write_bytes( |
| (recipe / 'document-boundaries.jsonl').read_bytes()) |
| (output / 'manifest.json').write_text( |
| json.dumps(manifest, ensure_ascii=False, indent=2) + '\n', encoding='utf-8') |
| write_checksums(output) |
| return {'case_count': len(cases), 'matched_source_and_reference_hashes': len(cases) * 2} |
|
|
|
|
| if __name__ == '__main__': |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument('--recipe', type=Path, default=Path(__file__).resolve().parent) |
| parser.add_argument('--source', type=Path, required=True, help='Official V1 Train train.zh') |
| parser.add_argument('--reference', type=Path, required=True, help='Official V1 Train train.en') |
| parser.add_argument('--output', type=Path, required=True, help='New output directory') |
| args = parser.parse_args() |
| print(json.dumps(reconstruct(args.recipe, args.source, args.reference, args.output))) |
|
|
