README.md

metadata

dataset_info:
  features:
    - name: messages
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
    - name: ground_truth
      dtype: string
    - name: dataset
      dtype: string
  splits:
    - name: train
      num_bytes: 18898829
      num_examples: 7473
    - name: test
      num_bytes: 3342727
      num_examples: 1319
  download_size: 2939167
  dataset_size: 22241556
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
license: mit

GSM8k Data - RLVR Formatted

This dataset contains the GSM8k dataset formatted for use with open-instruct - specifically reinforcement learning with verifiable rewards.

Part of the Tulu 3 release, for which you can see models here and datasets here.

Dataset Structure

Each example in the dataset contains the standard instruction-tuning data points as follow:

• messages (list): inputs used to prompt the model (after chat template formatting).
• ground_truth (str): the answer for the given sample.
• dataset (str): the name of the dataset, which determines which verifiable function is used.