Spaces:
Running
Running
""" | |
Train our Tokenizers on some data, just to see them in action. | |
The whole thing runs in ~25 seconds on my laptop. | |
""" | |
import os | |
import time | |
from minbpe import BasicTokenizer, RegexTokenizer | |
# open some text and train a vocab of 512 tokens | |
text = open("tests/taylorswift.txt", "r", encoding="utf-8").read() | |
# create a directory for models, so we don't pollute the current directory | |
os.makedirs("models", exist_ok=True) | |
t0 = time.time() | |
for TokenizerClass, name in zip([BasicTokenizer, RegexTokenizer], ["basic", "regex"]): | |
# construct the Tokenizer object and kick off verbose training | |
tokenizer = TokenizerClass() | |
tokenizer.train(text, 512, verbose=True) | |
# writes two files in the models directory: name.model, and name.vocab | |
prefix = os.path.join("models", name) | |
tokenizer.save(prefix) | |
t1 = time.time() | |
print(f"Training took {t1 - t0:.2f} seconds") |