metadata
dataset_info:
features:
- name: text
dtype: string
- name: meta
struct:
- name: redpajama_set_name
dtype: string
splits:
- name: train
num_bytes: 0
num_examples: 590394625
- name: validation
num_bytes: 0
num_examples: 502556
- name: test
num_bytes: 0
num_examples: 502268
download_size: 1507040721942
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
As datasets puts limits on the number of calls to huggingface, downloading SlimPajama-627B is problematic as it's composed of a ton of small files. I have reuploaded it here as larger chunks to easily download the dataset without having to do anything hacky.
The original dataset can be found here https://huggingface.co/datasets/cerebras/SlimPajama-627B