metadata
dataset_info:
features:
- name: questions
struct:
- name: id
sequence: int64
- name: text
sequence: string
- name: is_duplicate
dtype: bool
splits:
- name: train
num_bytes: 39231843
num_examples: 261317
- name: test
num_bytes: 7005599
num_examples: 44635
- name: val
num_bytes: 6704734
num_examples: 42232
download_size: 31031925
dataset_size: 52942176
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
- split: val
path: data/val-*
This is a cleaned version of the Quora dataset that's been configured with a train-test-val split.
- Train : For training model
- Test : For running experiments and comparing different OSS models and closed sourced models
- Val : Only to be used at the end!
Colab Notebook to reproduce : https://colab.research.google.com/drive/1dGjGiqwPV1M7JOLfcPEsSh3SC37urItS?usp=sharing