metadata
dataset_info:
features:
- name: inputs
dtype: string
- name: targets
dtype: string
- name: task
dtype: string
splits:
- name: train
num_bytes: 20917402.4513315
num_examples: 12635
download_size: 17409381
dataset_size: 20917402.4513315
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Flan-v2 filtered
This dataset is a version of the original Flanv2 but much reduced
- only CoT task few-shot samples are here, both opt and noopt
- samples with <100 tokens in response have been removed