File size: 3,897 Bytes
257ea84 | 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 | """Restore the byte-identical native training cache from the dataset release.
Usage: python restore_native_cache.py DATASET_ROOT/five_source_train/native_cache NEW_DIRECTORY
Only the Python standard library is required. No model, tokenizer, or GPU is loaded.
"""
from array import array
import argparse
import gzip
import hashlib
import json
from pathlib import Path
import shutil
import sys
def sha(path):
with Path(path).open('rb') as stream:
result=hashlib.sha256()
for chunk in iter(lambda:stream.read(4*1024*1024),b''):result.update(chunk)
return result.hexdigest()
def line(stream,value):
stream.write(json.dumps(value,ensure_ascii=False,separators=(',',':'),allow_nan=False)+'\n')
def main():
parser=argparse.ArgumentParser(description=__doc__)
parser.add_argument('cache',type=Path);parser.add_argument('output',type=Path)
args=parser.parse_args();root=args.cache.resolve();out=args.output.resolve()
if out.exists():raise FileExistsError('choose a fresh output directory')
manifest=json.loads((root/'CACHE_MANIFEST.json').read_text())
for item in manifest['compressed_files']:
p=(root/item['path']).resolve()
if not p.is_relative_to(root) or p.stat().st_size!=item['bytes'] or sha(p)!=item['sha256']:
raise ValueError('compressed cache file identity mismatch')
out.mkdir(parents=True)
for item in manifest['compressed_files']:
target=out/item['restored_file']
mode='ab' if item.get('append') else 'wb'
with gzip.open(root/item['path'],'rb') as source,target.open(mode) as dest:
shutil.copyfileobj(source,dest,length=4*1024*1024)
for name in ['COMPLETE.json','data_manifest.json','calibration_selection.json']:
shutil.copyfile(root/name,out/name)
original=json.loads((out/'data_manifest.json').read_text())
token_stream=(out/'tokens.uint32').open('rb')
def tokens(entry):
token_stream.seek(entry['offset_tokens']*4)
data=token_stream.read(entry['token_count']*4)
if len(data)!=entry['token_count']*4 or hashlib.sha256(data).hexdigest()!=entry['token_bytes_sha256']:
raise ValueError('per-record token bytes differ')
values=array('I');values.frombytes(data)
if sys.byteorder!='little':values.byteswap()
return values.tolist()
fingerprint=original['tokenizer']['fingerprint']
with (out/'document_index.jsonl').open(encoding='utf-8') as source,(out/'train_validation.jsonl').open('w',encoding='utf-8',newline='\n') as dest:
for raw in source:
entry=json.loads(raw)
value={k:entry[k] for k in ['id','group_id','source','split']}
value.update(tokenizer_identity=fingerprint,token_ids=tokens(entry),metadata=entry['metadata'])
line(dest,value)
with (out/'event_index.jsonl').open(encoding='utf-8') as source,(out/'train_validation_event_streams.jsonl').open('w',encoding='utf-8',newline='\n') as dest:
for raw in source:
entry=json.loads(raw)
value={k:entry[k] for k in ['id','group_id','source','split','tokenizer_identity','metadata']}
value['articles']=[{'id':a['id'],'token_ids':tokens(a),'metadata':a['metadata']} for a in entry['articles']]
line(dest,value)
token_stream.close()
for name,expected in original['files'].items():
path=out/name
if path.stat().st_size!=expected['bytes'] or sha(path)!=expected['sha256']:
raise ValueError('restored historical cache differs: '+name)
print(json.dumps({'status':'pass','restored_files':len(original['files'])+2,
'complete_sha256':sha(out/'COMPLETE.json'),
'data_manifest_sha256':sha(out/'data_manifest.json'),
'expanded_token_JSONL_reconstructed_byte_identically':True}))
if __name__=='__main__':main()
|