metadata
dataset_info:
features:
- name: prompt
dtype: string
- name: target
dtype: int64
- name: nums
list: int64
- name: ground_truth
struct:
- name: numbers
list: int64
- name: target
dtype: int64
- name: mean_reward
dtype: float64
- name: pass_rate
dtype: float64
- name: format_rate
dtype: float64
- name: reward_std
dtype: float64
- name: max_reward
dtype: float64
- name: mean_logprob
dtype: float64
- name: logprob_std
dtype: float64
- name: mean_logprob_correct
dtype: float64
- name: mean_logprob_incorrect
dtype: float64
- name: mean_response_length
dtype: float64
- name: response_length_std
dtype: float64
- name: unique_answers
dtype: int64
splits:
- name: train
num_bytes: 15571056
num_examples: 18432
download_size: 2061190
dataset_size: 15571056
configs:
- config_name: default
data_files:
- split: train
path: data/train-*