Skip to content

Dataset · markov-ai

CAD Tasks with Video and Reference Files

markov-ai/cad-1000-hours

509 end-to-end CAD tasks across seven programs: video, input and output files, and evaluation criteria.

What's inside

The dataset contains 509 end-to-end Windows single-screen CAD task execution recordings for seven CAD program families. Each task has task_desc.json with a prompt, application, input file paths, and expected results, input_files and output_files folders, rubrics.json with evaluation criteria, and task_overview.pdf. Total volume is 1,018.1229 hours; modalities are documents and video.

What you can do with it

  • Train a VLM agent to control a CAD interface using video
  • Evaluate an agent on end-to-end CAD tasks using rubrics.json
  • Create LoRA for understanding CAD screenshots
Downloads this month
105.4K
Likes
612

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("markov-ai/cad-1000-hours")