fw-45b / README.md
aklein4's picture
Update README.md
c9497cb verified
|
raw
history blame
467 Bytes
metadata
license: mit
task_categories:
  - text-generation
language:
  - en
pretty_name: fineweb-sample-100BT-tokenized-45b
size_categories:
  - 10B<n<100B

Fineweb sample-100BT tokenized with the gpt2 tokenizer. Sequences are truncated to a length of 1024 tokens.

For efficiency, tokens are stored in uint16 numpy arrays. Shards are stored in the webdataset format.