Datasets:

hbXNov
/

dynamath_mini

dataset_info:
  features:
    - name: id
      dtype: string
    - name: question
      dtype: string
    - name: image
      dtype: string
    - name: decoded_image
      dtype: image
    - name: ground_truth
      dtype: string
    - name: answer_type
      dtype: string
    - name: subject
      dtype: string
    - name: knowledge_level
      dtype: string
  splits:
    - name: train
      num_bytes: 37409713.77245509
      num_examples: 1000
  download_size: 43769777
  dataset_size: 37409713.77245509
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

(Unofficial) Mini version of Dynamath (https://huggingface.co/datasets/DynaMath/DynaMath_Sample) dataset.

Here, we have sampled 100 random ids from all 10 variants of the original Dynamath dataset.

This allows faster inference on 1000 eval examples instead of 5010 examples; which can be very slow and limiting for large model runs.