README.md

---
dataset_info:
- config_name: testmini
  features:
  - name: id
    dtype: string
  - name: media
    list: image
  - name: messages
    dtype: string
  - name: answer
    dtype: string
  - name: question_type
    dtype: string
  - name: source_question_type
    dtype: string
  - name: problem_version
    dtype: string
  - name: subject
    dtype: string
  - name: subfield
    dtype: string
  - name: source
    dtype: string
  splits:
  - name: testmini
    num_bytes: 202954938
    num_examples: 3940
  download_size: 201549689
  dataset_size: 202954938
- config_name: testmini_text_only
  features:
  - name: id
    dtype: string
  - name: media
    list: image
  - name: messages
    dtype: string
  - name: answer
    dtype: string
  - name: question_type
    dtype: string
  - name: source_question_type
    dtype: string
  - name: problem_version
    dtype: string
  - name: subject
    dtype: string
  - name: subfield
    dtype: string
  - name: source
    dtype: string
  splits:
  - name: testmini_text_only
    num_bytes: 388102
    num_examples: 788
  download_size: 100391
  dataset_size: 388102
configs:
- config_name: testmini
  data_files:
  - split: testmini
    path: testmini/testmini-*
- config_name: testmini_text_only
  data_files:
  - split: testmini_text_only
    path: testmini_text_only/testmini_text_only-*
---

# MathVerse (mm-eval mirror)

Mirror of [AI4Math/MathVerse](https://huggingface.co/datasets/AI4Math/MathVerse) in Simple-MMEval format.

- Prompt template source (Tier 2): the source dataset's `query_wo` column (pre-rendered single-answer evaluation prompt from the official eval pipeline). Template: `<image>{{ question }}`.
- See `metadata.json` at the repo root for the full per-subset manifest.