versae's picture
Adjust batch size for extrating tokens
8b9ba87
899 Bytes
#!/usr/bin/env python3
from datasets import load_dataset
from tokenizers import ByteLevelBPETokenizer
# Load dataset
<<<<<<< HEAD
dataset = load_dataset("oscar", "unshuffled_deduplicated_es", split="train[:5000000]")
# Instantiate tokenizer
tokenizer = ByteLevelBPETokenizer()
def batch_iterator(batch_size=100_000):
=======
dataset = load_dataset("oscar", "unshuffled_deduplicated_es", split="train")
# Instantiate tokenizer
tokenizer = ByteLevelBPETokenizer()
def batch_iterator(batch_size=1_000_000):
>>>>>>> d5cede47e74aa6ec36f20acf5aba37c6734c6186
for i in range(0, len(dataset), batch_size):
yield dataset["text"][i: i + batch_size]
# Customized training
tokenizer.train_from_iterator(batch_iterator(), vocab_size=50265, min_frequency=2, special_tokens=[
"<s>",
"<pad>",
"</s>",
"<unk>",
"<mask>",
])
# Save files to disk
tokenizer.save("./tokenizer.json")