File size: 1,060 Bytes
38e3c7e 5feb9da 155f524 5feb9da |
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 |
---
dataset_info:
features:
- name: input_ids
sequence: int32
splits:
- name: train
num_bytes: 35168157552.0
num_examples: 1073116
download_size: 18303136476
dataset_size: 35168157552.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# OpenWebTextCorpus tokenized for Gemma
This dataset is a pre-tokenized version of the [Skylion007/openwebtext](https://huggingface.co/datasets/Skylion007/openwebtext) dataset
using the [gemma](https://huggingface.co/google/gemma-2b) tokenizer. As such, this dataset follows the same licensing as the original openwebtext dataset.
This pre-tokenization is done as a performance optimization for using the openwebtext dataset with a Gemma model (gemma-2b, gemma-2b-it, gemma-7b, gemma-7b-it).
This dataset was created using [SAELens](https://github.com/jbloomAus/SAELens), with the following settings:
- context_size: 8192
- shuffled: true
- begin_batch_token: "bos"
- begin_sequence_token: null
- sequence_separator_token: "bos"
- sae_lens_version: "3.3.0" |