pro-shant/

textcad-bench

Frozen evaluation dataset for TextCAD-Bench — a unified benchmark for LLM-based text-to-CAD generation.

	Files




	

File Samples Description

t1_samples.json 2,000 DeepCAD test set, L1 prompts, stratified by complexity

t2_samples.json 1,725 Fusion360 industrial parts, GPT-4.1 annotated

t3_cadprompt.json 200 Human-authored CadQuery-style prompts

gt_meshes.zip — Ground-truth OBJ meshes for T1

gt_meshes_t2.zip — Ground-truth OBJ meshes for… See the full description on the dataset page: https://huggingface.co/datasets/pro-shant/textcad-bench.