README.md

metadata

configs:
  - config_name: edit-bench
    data_files:
      - split: edit_bench
        path: edit-bench/data/edit_bench-*.parquet
  - config_name: code_gen
    data_files:
      - split: code_gen
        path: code_gen/data/code_gen-*.parquet
  - config_name: QA
    data_files:
      - split: QA
        path: QA/qa_*.parquet
license: apache-2.0
task_categories:
  - image-to-text
  - text-generation
  - question-answering
size_categories:
  - 10K<n<100K
tags:
  - cad
  - cadquery
  - 3d
  - synthetic
  - code-generation

BenchCAD

Three-config dataset for CAD evaluation:

• edit-bench — held-out CAD edit benchmark.
• code_gen — 17,900 synthetic CadQuery samples (compact 12-column variant) covering 106 mechanical part families. Each row contains the GT CadQuery code plus 5 normalized renders.
• QA — CAD question-answering benchmark.

code_gen schema (12 columns)

Column	Type	Description
stem	string	unique sample identifier
family	string	mechanical part family (106 distinct)
variant	string	sub-variant within family
difficulty	string	easy / medium / hard
base_plane	string	initial workplane (XY / XZ / YZ)
standard	string	ISO/DIN standard if applicable
code	string	CadQuery Python source (ground truth)
view_0_png	image	front view (134×134 PNG)
view_1_png	image	right view
view_2_png	image	top view
view_3_png	image	iso view
composite_png	image	2×2 composite of the four views

Usage

from datasets import load_dataset

ds = load_dataset("BenchCAD/BenchCAD", "code_gen", split="code_gen")
print(ds[0]["family"], ds[0]["difficulty"])
ds[0]["composite_png"].show()
print(ds[0]["code"])