File size: 3,072 Bytes
569cec7 8092987 569cec7 | 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 | #!/usr/bin/env python3
"""Generate or check fast-tokenizer references at the revisions in manifest.json.
Adapted from apocryphx/swift-transformers PR #360; see README.md for attribution.
Only Python reference outputs are written; Swift output never becomes a golden.
"""
import argparse
import datetime
import json
from pathlib import Path
import tokenizers
import transformers
from transformers import AutoTokenizer
ROOT = Path(__file__).resolve().parent.parent
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--check', action='store_true', help='Compare reference entries without writing files')
args = parser.parse_args()
manifest = json.loads((ROOT / 'manifest.json').read_text(encoding='utf-8'))
failed = False
for corpus in manifest['corpora']:
inputs = json.loads((ROOT / corpus['inputs']).read_text(encoding='utf-8'))
assert inputs and len({i['id'] for i in inputs}) == len(inputs), 'Empty corpus or duplicate IDs'
for spec in corpus['baselines']:
model = spec['model_id']
revision = spec['revision']
assert len(revision) == 40 and all(c in '0123456789abcdef' for c in revision)
print(f'{corpus["id"]}: {model}@{revision}', flush=True)
tokenizer = AutoTokenizer.from_pretrained(model, revision=revision, use_fast=True)
assert tokenizer.is_fast, 'References must use the fast tokenizer'
entries = []
for row in inputs:
ids = tokenizer(row['text'], add_special_tokens=True)['input_ids']
entries.append({
'id': row['id'],
'input_ids': ids,
'tokens': tokenizer.convert_ids_to_tokens(ids),
'decoded_with_special': tokenizer.decode(ids, skip_special_tokens=False),
'decoded_skip_special': tokenizer.decode(ids, skip_special_tokens=True),
})
path = ROOT / spec['path']
if args.check:
previous = json.loads(path.read_text(encoding='utf-8'))['entries']
matches = previous == entries
print(f' {len(entries)} entries: {"MATCH" if matches else "DIFFER"}', flush=True)
failed |= not matches
continue
payload = {
'metadata': {
'model_id': model, 'model_revision': revision,
'transformers_version': transformers.__version__,
'tokenizers_version': tokenizers.__version__,
'generated_at': datetime.datetime.now(datetime.timezone.utc).replace(microsecond=0).isoformat(),
'input_count': len(entries),
},
'entries': entries,
}
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + '\n', encoding='utf-8')
raise SystemExit(1 if failed else 0)
if __name__ == '__main__':
main()
|