metadata
dataset_info:
features:
- name: model_inputs
struct:
- name: preorder_index
sequence: int32
- name: postorder_index
sequence: int32
- name: input_ids
sequence: int32
- name: labels
sequence: int32
- name: position_ids
sequence: int32
- name: sequence_id
sequence: int32
- name: loss_inputs
struct:
- name: parent_index
sequence: int32
- name: num_correct
sequence: int32
- name: num_rollouts
sequence: int32
- name: parent_rollout_index
sequence: int32
splits:
- name: train
num_bytes: 548713552
num_examples: 6636
download_size: 59547816
dataset_size: 548713552
configs:
- config_name: default
data_files:
- split: train
path: data/train-*