Created
March 27, 2021 10:28
-
-
Save kingoflolz/2c63f01f388ae2706b5d395b5942e63b to your computer and use it in GitHub Desktop.
A quick script for shuffling tfrecord datasets
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import tensorflow as tf | |
from tqdm import tqdm | |
index = open("data/openwebtext2_new_inputs.train.index").read().splitlines() | |
dataset = tf.data.Dataset.from_tensor_slices(index) | |
dataset = dataset.interleave(tf.data.TFRecordDataset, cycle_length=128, num_parallel_calls=tf.data.experimental.AUTOTUNE) | |
d = dataset.shuffle(10000).prefetch(100) | |
i = 0 | |
for idx, example in enumerate(tqdm(d)): | |
if idx % 100000 == 0: | |
try: | |
writer.close() | |
except: | |
pass | |
writer = tf.io.TFRecordWriter(f"gs://neo-datasets/openwebtext2_new_inputs_shuffled/train/openwebtext_{i}.tfrecords") | |
i += 1 | |
writer.write(example.numpy()) | |
writer.close() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment