Datasets:
File size: 2,995 Bytes
9379442 | 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 81 82 83 84 85 86 87 88 | """Small, deterministic helpers for the four dataset configurations."""
from __future__ import annotations
import hashlib
import json
from pathlib import Path
import pyarrow as pa
import pyarrow.parquet as pq
ROOT = Path(__file__).resolve().parents[1]
def load_json(path):
return json.loads(Path(path).read_text(encoding='utf-8'))
def write_json(path, value, *, compact=False):
path = Path(path)
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(json.dumps(value, ensure_ascii=False, indent=None if compact else 2,
separators=(',', ':') if compact else None) + '\n', encoding='utf-8')
def sha256(path):
result = hashlib.sha256()
with Path(path).open('rb') as handle:
for block in iter(lambda: handle.read(4194304), b''):
result.update(block)
return result.hexdigest()
def id_hash(ids):
return hashlib.sha256(('\n'.join(sorted(ids)) + '\n').encode()).hexdigest()
def normalized_prompt(text):
return ' '.join(text.split())
def ids_from_table(table):
if 'problem_id' in table.column_names:
return table['problem_id'].to_pylist()
return [row['problem_id'] for row in table['extra_info'].to_pylist()]
def write_parquet(path, table):
path = Path(path)
path.parent.mkdir(parents=True, exist_ok=True)
pq.write_table(table, path, compression='zstd', row_group_size=1024)
def describe_file(path, root):
path = Path(path)
parquet = pq.ParquetFile(path)
columns = ['problem_id'] if 'problem_id' in parquet.schema_arrow.names else ['extra_info']
ids = ids_from_table(parquet.read(columns=columns))
return {'path': path.relative_to(root).as_posix(), 'rows': parquet.metadata.num_rows,
'bytes': path.stat().st_size, 'sha256': sha256(path),
'problem_id_set_sha256': id_hash(ids), 'columns': parquet.schema_arrow.names}
def make_manifest(root, family, split_policy):
files = {}
for kind in ['sft', 'rl']:
for split in ['train', 'validation']:
path = root / 'data' / family / kind / (split + '.parquet')
files[f'{kind}/{split}'] = describe_file(path, root)
result = {'version': '1.0.0', 'family': family, 'split_policy': split_policy, 'files': files}
write_json(root / 'metadata' / family / 'dataset_manifest.json', result)
return result
def member_record(ids, removed_category=None):
return {'rows': len(ids), 'problem_ids': ids, 'problem_id_set_sha256': id_hash(ids),
'removed_category': removed_category}
def export_rows(table, ids, destination):
all_ids = ids_from_table(table)
lookup = {v: i for i, v in enumerate(all_ids)}
if len(lookup) != len(all_ids):
raise ValueError('Duplicate source IDs')
missing = set(ids) - lookup.keys()
if missing:
raise ValueError(f'{len(missing)} arm IDs absent from this split')
write_parquet(destination, table.take(pa.array([lookup[v] for v in ids], type=pa.int64())))
|