---
configs:
- config_name: algorithm
  data_files:
  - split: train
    path: algorithm/train-*.parquet
- config_name: demo-mm
  data_files:
  - split: train
    path: demo-mm/train-*.parquet
- config_name: demo-text
  data_files:
  - split: train
    path: demo-text/train-*.parquet
- config_name: html-edit
  data_files:
  - split: train
    path: html-edit/train-*.parquet
- config_name: html-gen
  data_files:
  - split: train
    path: html-gen/train-*.parquet
- config_name: manim
  data_files:
  - split: train
    path: manim/train-*.parquet
- config_name: python-mm
  data_files:
  - split: train
    path: python-mm/train-*.parquet
- config_name: python-text
  data_files:
  - split: train
    path: python-text/train-*.parquet
- config_name: chemistry
  data_files:
  - split: train
    path: chemistry/train-*.parquet
- config_name: asymptote
  data_files:
  - split: train
    path: asymptote/train-*.parquet
- config_name: latex
  data_files:
  - split: train
    path: latex/train-*.parquet
- config_name: lilypond
  data_files:
  - split: train
    path: lilypond/train-*.parquet
- config_name: mermaid
  data_files:
  - split: train
    path: mermaid/train-*.parquet
- config_name: svg
  data_files:
  - split: train
    path: svg/train-*.parquet
- config_name: vegalite
  data_files:
  - split: train
    path: vegalite/train-*.parquet
- config_name: javascript
  data_files:
  - split: train
    path: javascript/train-*.parquet
task_categories:
- image-text-to-text
---

# PrismaCoder-2M

PrismaCoder-2M is a large-scale multimodal code intelligence dataset introduced in the paper [PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis](https://huggingface.co/papers/2610.20538).

- **Project page:** https://qiushisun.github.io/PrismaCoder-Home/
- **Code:** https://github.com/QiushiSun/PrismaCoder

The dataset contains 2M examples covering chart generation, web interfaces, interactive demos, vector graphics, scientific documents, chemical structures, and animations. It includes both vision-centric and text-centric tasks.

We are still actively uploading the dataset. The full dataset will be available soon.