Datasets:
File size: 11,879 Bytes
afecc1c | 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 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 | """
UKS Seed Dataset Generator
Generates preprocessed training data from FineWeb dataset using:
- spaCy for linguistic features (POS, dependencies, lemmas, entities)
- XLNetTokenizer for subword tokenization
Outputs:
- pre_tokenized.jsonl.gz: Ready for model training
- intermediate.parquet: For exploration and custom tokenization
"""
import gzip
import hashlib
import json
from dataclasses import dataclass, field
from pathlib import Path
from typing import Dict, Iterator, List, Optional
import spacy
from datasets import load_dataset
from transformers import XLNetTokenizer
from tqdm import tqdm
@dataclass
class SEEDConfig:
max_samples: int = 10000
batch_size: int = 100
min_sentence_len: int = 5
max_sentence_len: int = 512
output_dir: Path = Path(__file__).parent / "train"
cache_dir: Path = Path(__file__).parent / ".cache"
hf_dataset_name: str = "HuggingFaceFW/fineweb"
hf_dataset_split: str = "train"
spacy_model: str = "en_core_web_sm"
xlnet_model: str = "xlnet/xlnet-base-cased"
class SEEDDatasetGenerator:
def __init__(self, config: Optional[SEEDConfig] = None):
self.config = config or SEEDConfig()
self._nlp = None
self._tokenizer = None
self._processed_ids: set = set()
@property
def nlp(self):
if self._nlp is None:
try:
self._nlp = spacy.load(self.config.spacy_model)
except OSError:
import subprocess
import sys
print(f"Downloading spaCy model: {self.config.spacy_model}")
subprocess.run(
[
sys.executable,
"-m",
"spacy",
"download",
self.config.spacy_model,
],
check=True,
)
self._nlp = spacy.load(self.config.spacy_model)
return self._nlp
@property
def tokenizer(self):
if self._tokenizer is None:
self._tokenizer = XLNetTokenizer.from_pretrained(self.config.xlnet_model)
return self._tokenizer
def _load_processed_ids(self) -> set:
cache_file = self.config.cache_dir / "processed_ids.json"
if cache_file.exists():
with open(cache_file, "r") as f:
return set(json.load(f))
return set()
def _save_processed_ids(self):
cache_file = self.config.cache_dir / "processed_ids.json"
cache_file.parent.mkdir(parents=True, exist_ok=True)
with open(cache_file, "w") as f:
json.dump(list(self._processed_ids), f)
def _get_sample_id(self, text: str) -> str:
return hashlib.sha256(text.encode()).hexdigest()[:16]
def _load_fineweb(self) -> Iterator[Dict]:
dataset = load_dataset(
self.config.hf_dataset_name,
split=self.config.hf_dataset_split,
streaming=True,
)
for sample in dataset:
text = sample.get("text", "")
if text and len(text.strip()) > 0:
yield {
"id": sample.get("id", ""),
"text": text,
}
def _process_document(self, text: str) -> List[Dict]:
doc = self.nlp(text)
records = []
for sent in doc.sents:
sent_text = sent.text.strip()
if len(sent_text) < self.config.min_sentence_len:
continue
if len(sent_text) > self.config.max_sentence_len:
sent_text = sent_text[: self.config.max_sentence_len]
tokens = []
pos_tags = []
dep_tags = []
lemmas = []
ent_types = []
head_dists = []
for token in sent:
tokens.append(token.text)
pos_tags.append(token.pos_)
dep_tags.append(token.dep_)
lemmas.append(token.lemma_ if token.lemma_ else token.text)
ent_types.append(token.ent_type_ if token.ent_type_ else "O")
head_dists.append(token.head.i - token.i)
enc = self.tokenizer(
sent_text,
add_special_tokens=False,
truncation=True,
max_length=self.config.max_sentence_len,
)
token_ids = enc["input_ids"]
record = {
"id": self._get_sample_id(sent_text),
"text": sent_text,
"tokens": json.dumps(tokens),
"token_ids": json.dumps(token_ids),
"pos": json.dumps(pos_tags),
"dep": json.dumps(dep_tags),
"lemma": json.dumps(lemmas),
"ent_type": json.dumps(ent_types),
"head_dist": json.dumps(head_dists),
}
records.append(record)
return records
def generate(self) -> Dict[str, Path]:
self.config.output_dir.mkdir(parents=True, exist_ok=True)
self.config.cache_dir.mkdir(parents=True, exist_ok=True)
self._processed_ids = self._load_processed_ids()
jsonl_path = self.config.output_dir / "pre_tokenized.jsonl.gz"
parquet_path = self.config.output_dir / "intermediate.parquet"
jsonl_records: List[Dict] = []
all_records: List[Dict] = []
try:
import pyarrow as pa
import pyarrow.parquet as pq
has_pyarrow = True
except ImportError:
has_pyarrow = False
print("Warning: pyarrow not installed. Parquet output will be skipped.")
print("Install with: pip install pyarrow")
batch_buffer: List[str] = []
sample_count = 0
skipped_count = 0
print(f"Loading FineWeb from: {self.config.hf_dataset_name}")
print(f"Target samples: {self.config.max_samples}")
dataset = self._load_fineweb()
for sample in tqdm(dataset, desc="Processing documents"):
if sample_count >= self.config.max_samples:
break
sample_text = sample["text"]
if not sample_text:
continue
batch_buffer.append(sample_text)
if len(batch_buffer) >= self.config.batch_size:
texts = batch_buffer
batch_buffer = []
docs = list(self.nlp.pipe(texts, batch_size=len(texts)))
for doc in docs:
for sent in doc.sents:
sent_text = sent.text.strip()
if len(sent_text) < self.config.min_sentence_len:
continue
if len(sent_text) > self.config.max_sentence_len:
sent_text = sent_text[: self.config.max_sentence_len]
sample_id = self._get_sample_id(sent_text)
if sample_id in self._processed_ids:
skipped_count += 1
continue
self._processed_ids.add(sample_id)
sample_count += 1
tokens = []
pos_tags = []
dep_tags = []
lemmas = []
ent_types = []
head_dists = []
for token in sent:
tokens.append(token.text)
pos_tags.append(token.pos_)
dep_tags.append(token.dep_)
lemmas.append(token.lemma_ if token.lemma_ else token.text)
ent_types.append(
token.ent_type_ if token.ent_type_ else "O"
)
head_dists.append(token.head.i - token.i)
enc = self.tokenizer(
sent_text,
add_special_tokens=False,
truncation=True,
max_length=self.config.max_sentence_len,
)
token_ids = enc["input_ids"]
record = {
"id": sample_id,
"text": sent_text,
"tokens": json.dumps(tokens),
"token_ids": json.dumps(token_ids),
"pos": json.dumps(pos_tags),
"dep": json.dumps(dep_tags),
"lemma": json.dumps(lemmas),
"ent_type": json.dumps(ent_types),
"head_dist": json.dumps(head_dists),
}
jsonl_records.append(record)
all_records.append(record)
if sample_count >= self.config.max_samples:
break
if sample_count >= self.config.max_samples:
break
if sample_count % 1000 == 0 and jsonl_records:
self._write_jsonl(jsonl_records, jsonl_path, append=True)
jsonl_records = []
self._save_processed_ids()
if jsonl_records:
self._write_jsonl(jsonl_records, jsonl_path, append=True)
if has_pyarrow and all_records:
self._write_parquet(all_records, parquet_path)
self._save_processed_ids()
print(f"\n=== Generation Complete ===")
print(f"Total samples: {sample_count}")
print(f"Skipped (duplicates): {skipped_count}")
print(f"Output files:")
print(f" - {jsonl_path}")
if has_pyarrow:
print(f" - {parquet_path}")
return {
"jsonl": jsonl_path,
"parquet": parquet_path if has_pyarrow else None,
}
def _write_jsonl(self, records: List[Dict], path: Path, append: bool = False):
mode = "ab" if append else "wb"
with gzip.open(path, mode) as f:
for record in records:
f.write((json.dumps(record) + "\n").encode())
def _write_parquet(self, records: List[Dict], path: Path):
try:
import pyarrow as pa
import pyarrow.parquet as pq
if not records:
return
keys = list(records[0].keys())
arrays = []
for key in keys:
arrays.append(
pa.array([r.get(key, "") for r in records], type=pa.string())
)
schema = pa.schema(zip(keys, [pa.string() for _ in keys]))
table = pa.table(dict(zip(keys, arrays)), schema=schema)
pq.write_table(table, path)
except ImportError:
print("Skipping parquet output (pyarrow not available)")
def main():
import argparse
parser = argparse.ArgumentParser(description="Generate UKS Seed Dataset")
parser.add_argument(
"--max-samples", type=int, default=10000, help="Maximum number of samples"
)
parser.add_argument(
"--batch-size", type=int, default=100, help="Batch size for processing"
)
parser.add_argument("--output-dir", type=str, default=None, help="Output directory")
parser.add_argument("--cache-dir", type=str, default=None, help="Cache directory")
args = parser.parse_args()
config = SEEDConfig(
max_samples=args.max_samples,
batch_size=args.batch_size,
)
if args.output_dir:
config.output_dir = Path(args.output_dir)
if args.cache_dir:
config.cache_dir = Path(args.cache_dir)
generator = SEEDDatasetGenerator(config)
generator.generate()
if __name__ == "__main__":
main()
|