datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
droid_1.0.1This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "Franka",
"total_episodes": 95600,
"total_frames": 27612581,
"total_tasks": 0,
"total_videos": 286800,
"total_chunks": 95,
"chunks_size": 1000,
"fps": 15,
"splits": {
"train": "0:95600"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/cadene/droid_1.0.1.cad-1000-hours
CAD-1K Open v2 - 1,018.1229 Hours
509 end-to-end, single-display Windows CAD task recordings across seven CAD software families.
Each task contains:
task_desc.json - task prompt, application, reference-input paths, and expected deliverables
input_files/ - reference inputs named input.ext or input_N.ext
output_files/ - submitted CAD deliverables and supplemental outputs named output.ext or output_N.ext
rubrics.json - task-specific evaluation criteria
task_overview.pdf - review… See the full description on the dataset page: https://huggingface.co/datasets/markov-ai/cad-1000-hours.droidThis dataset was created using LeRobot.
DROID: A Large-Scale In-the-Wild Robot Manipulation Dataset
One of the biggest open-source dataset for robotics with 27.044,326 frames, 92,223 episodes, 31,308 unique task description in natural language.
Ported from Tensorflow Dataset format (2TB) to LeRobotDataset format (400GB) with the help from IPEC-COMMUNITY.
Visualization: LeRobot
Homepage: Droid
Paper: Arxiv
License: apache-2.0
Dataset Structure
meta/info.json:
{… See the full description on the dataset page: https://huggingface.co/datasets/cadene/droid.agibot_alpha_v30This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "AgiBot_A2D",
"total_episodes": 28122,
"total_frames": 47613574,
"total_tasks": 30,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:28122"},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/cadene/agibot_alpha_v30.cad-environments
CAD Environments
CAD Environments is a multimodal dataset of complete, human-performed workflows in desktop CAD software. The current release contains 51 task workflows totaling 99.03 hours, covering eight software groups across mechanical design, architecture, MEP, structural design, and general 3D modeling.
Each workflow preserves the full task context—not just the final model—including the problem statement, reference and input files, a gold output, evaluation rubrics, a… See the full description on the dataset page: https://huggingface.co/datasets/markov-ai/cad-environments.CADS-dataset
CADS: A Comprehensive Anatomical Dataset and Segmentation for Whole-Body Anatomy in Computed Tomography
Overview
CADS is a robust, fully automated framework for segmenting 167 anatomical structures in Computed Tomography (CT), spanning from head to knee regions across diverse anatomical systems.
The framework consists of two main components:
CADS-dataset:
22,022 CT volumes with complete annotations for 167 anatomical structures.
Most extensive whole-body CT dataset… See the full description on the dataset page: https://huggingface.co/datasets/huggingface/CADS-dataset.cadgenbench-submissions
CADGenBench Submissions
Submissions and evaluation results for the CADGenBench leaderboard.
Leaderboard Space: HuggingAI4Engineering/CADGenBench.
Benchmark code: github.com/huggingface/cadgenbench.
Fixture inputs: HuggingAI4Engineering/cadgenbench-data.
Ground truth: HuggingAI4Engineering/cadgenbench-data-gt (private).
Contents
Path
What it is
results.jsonl
One row per submitted + evaluated submission. The leaderboard table reads from this.… See the full description on the dataset page: https://huggingface.co/datasets/HuggingAI4Engineering/cadgenbench-submissions.cad-gen-freecad-bench
Parametric CAD Bench — results dataset
Run-by-run results for Parametric CAD Bench, a benchmark that
measures whether AI agents can author editable FreeCAD models from
natural-language part descriptions. 1000 rows, one per
(agent, model, task_id, trial) over the
gnucleus-ai/cad-bench@v1
task suite. The public leaderboard view of this data lives at
cadbench.ai.
What's in here
data/cad-bench-v1.parquet — the row table. Each row carries the
composite + sub-scores… See the full description on the dataset page: https://huggingface.co/datasets/gnucleus-ai/cad-gen-freecad-bench.cad-recode-v1.5
CAD-Recode: Reverse Engineering CAD Code from Point Clouds
CAD-Recode dataset is provided in form of Python (CadQuery) codes.
Train size is ~1M and validation size is ~1k.
CAD-Recode model and code are released at github https://github.com/filaPro/cad-recode.
And if you like it, give us a github 🌟.
Citation
If you find this work useful for your research, please cite our paper:
@misc{rukhovich2024cadrecode,
title={CAD-Recode: Reverse Engineering CAD Code from Point… See the full description on the dataset page: https://huggingface.co/datasets/filapro/cad-recode-v1.5.liars-bench
Liars' Bench
Liars' Bench is a benchmark for evaluating lie-detectors for language models. For details, see paper here.
It was generated by using...
four open-weight models: mistral-small-3.1-24b-instruct, llama-v3.3-70b-instruct, qwen-2.5-72b-instruct, gemma-3-27b-it
across seven datasets including a control dataset (Alpaca).
Our settings capture qualitatively different types of lies and vary along two dimensions:
the model's reason for lying
the object of belief targeted… See the full description on the dataset page: https://huggingface.co/datasets/Cadenza-Labs/liars-bench.cad-corpus-cleanedZero-To-CAD-1m
Zero-to-CAD 1M
One million executable, interpretable CAD construction sequences synthesized entirely without real-world data.
Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data
Mohammadmehdi Ataei, Farzaneh Askari, Kamal Rahimi Malekshan, Pradeep Kumar Jayaraman
Autodesk Research
Related Resources
Resource
Link
📄 Paper
Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale… See the full description on the dataset page: https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m.cadgenbench-data
CADGenBench (Inputs)
Public inputs for the CADGenBench benchmark, which measures how well AI
systems produce correct 3D mechanical parts as STEP files. This repository
holds the task inputs only; the ground truth is withheld in a separate private
repository so that the leaderboard's evaluation is the single source of truth.
Leaderboard Space: HuggingAI4Engineering/CADGenBench
Browse the tasks: the Tasks tab on the Space (thumbnails, search, generation/editing filter, per-task… See the full description on the dataset page: https://huggingface.co/datasets/HuggingAI4Engineering/cadgenbench-data.CADBench
CADBench
CADBench is a unified multimodal benchmark for AI-assisted CAD program generation, introduced in the paper CADBench: A Multimodal Benchmark for AI-Assisted CAD Program Generation.
The benchmark contains 18,000 evaluation samples spanning six benchmark families derived from DeepCAD, Fusion 360, ABC, MCB, and Objaverse. It is designed to measure progress in editable 3D reconstruction and multimodal CAD understanding.
Dataset Summary
CADBench supports… See the full description on the dataset page: https://huggingface.co/datasets/DeCoDELab/CADBench.Omni-CAD-Subset-Completeeccv2026-cad-challenge-data
ECCV 2026 CAD Challenge Data
This challenge is part of the workshop The Path to Manufacturing: Evolving
3D Generation to Intelligent Computer-Aided Design.
Workshop homepage: https://3dgen-cad-workshop.github.io/
Challenge submission Space: https://huggingface.co/spaces/jingwei-xu-00/eccv2026-cad-challenge
Dataset rendering and preparation code (only .step files are required): https://github.com/DavidXu-JJ/eccv2026-cad-challenge-data-render
This repository contains the public… See the full description on the dataset page: https://huggingface.co/datasets/jingwei-xu-00/eccv2026-cad-challenge-data.CADS-dataset
CADS: A Comprehensive Anatomical Dataset and Segmentation for Whole-Body Anatomy in Computed Tomography
Overview
CADS is a robust, fully automated framework for segmenting 167 anatomical structures in Computed Tomography (CT), spanning from head to knee regions across diverse anatomical systems.
The framework consists of two main components:
CADS-dataset:
22,022 CT volumes with complete annotations for 167 anatomical structures.
Most extensive whole-body CT dataset… See the full description on the dataset page: https://huggingface.co/datasets/sunghong/CADS-dataset.Zero-To-CAD-100k
Zero-to-CAD 100K
A curated subset of 100,000 geometrically diverse CAD construction sequences selected from Zero-to-CAD 1M.
Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data
Mohammadmehdi Ataei, Farzaneh Askari, Kamal Rahimi Malekshan, Pradeep Kumar Jayaraman
Autodesk Research
Related Resources
Resource
Link
📄 Paper
Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without… See the full description on the dataset page: https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-100k.cad-recode
CAD-Recode: Reverse Engineering CAD Code from Point Clouds
CAD-Recode dataset is provided in form of Python (CadQuery) codes.
Train size is ~1M and validation size is ~1k.
CAD-Recode model and code are released at github https://github.com/filaPro/cad-recode.
And if you like it, give us a github 🌟.
Citation
If you find this work useful for your research, please cite our paper:
@misc{rukhovich2024cadrecode,
title={CAD-Recode: Reverse Engineering CAD Code from Point… See the full description on the dataset page: https://huggingface.co/datasets/filapro/cad-recode.CADFS
CADFS Dataset
CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models
🚀 Project Page
📃 Paper
💻 Code
🤗 Model
A large-scale dataset for parametric CAD model generation from text descriptions and multi-view images. Models are represented as FeatureScript programs, enabling direct import into Onshape environment.
This dataset was used to train and evaluate CADFS-2B, a fine-tuned Qwen2-VL-2B multimodal language model… See the full description on the dataset page: https://huggingface.co/datasets/VladPyatov/CADFS.cad-gen-freecad
CAD Generation Dataset
Each row in this dataset describes one parametric CAD part. Columns:
id — row identifier (also the basename of the per-row asset files)
name — part family (e.g. flanges, spur_gear_stock)
description — natural-language description of the geometry
key_parameters — the dimensions that drive the parametric model
image — 512×512 PNG preview rendered from the FCStd
fcstd_path — relative path inside this repo to the parametric FreeCAD document (fcstd/<id>.FCStd)… See the full description on the dataset page: https://huggingface.co/datasets/gnucleus-ai/cad-gen-freecad.droid_1.0.1_v30This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "Franka",
"total_episodes": 95584,
"total_frames": 27607757,
"total_tasks": 49596,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 15,
"splits": {
"train": "0:95584"},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/cadene/droid_1.0.1_v30.cadif-followir-results
CADIF Results: De-confounded Instruction-Following Retrieval
Per-query results, contrasts, and run checkpoints for CADIF (a Composition Assay for
De-confounding Instruction-Following), a control-task battery for instruction-following
retrieval. A paper describing CADIF is in preparation.
CADIF asks whether an instruction-conditioned retriever actually uses an instruction's
composition, or only exploits the topical words it adds. The headline readout is a within-model
contrast… See the full description on the dataset page: https://huggingface.co/datasets/tardellirs/cadif-followir-results.cad-steps
CAD-Steps: Intermediate CAD Construction States Dataset
A large-scale dataset of intermediate CAD geometry states, capturing the step-by-step construction process of 3D models.
Overview
CAD-Steps provides paired data of CAD construction operations and their resulting geometry at each intermediate step. Unlike existing CAD datasets that only provide final geometry (ABC) or symbolic operations (DeepCAD), this dataset includes the actual 3D geometry at every construction… See the full description on the dataset page: https://huggingface.co/datasets/amzyst1/cad-steps.cad-gen-freecad-bench-v2
Parametric CAD Bench v2 — results dataset
Run-by-run results for Parametric CAD Bench v2, a benchmark that measures
whether AI agents can author editable FreeCAD models from natural-language part
descriptions. This archive contains 1,000 rows: one trial for each of 100 tasks
across the 10 public jobs on the live
gnucleus-ai/cad-bench@v2
leaderboard.
What's in here
data/cad-bench-v2.parquet — the trial index. Each row carries the
continuous reward and its… See the full description on the dataset page: https://huggingface.co/datasets/gnucleus-ai/cad-gen-freecad-bench-v2.cad-1000-hours
CAD 1000 Hours
CAD 1000 Hours is a computer-use dataset containing 1,021.64 hours of recorded work across 597 workflows and 10 CAD, BIM, structural-analysis, and visualization applications. The tables below summarize its software coverage.
Category distribution
Category
Included software
Workflows
Hours
Share of hours
Drafting and general CAD
AutoCAD
238
501.99
49.14%
Mechanical and product CAD
SOLIDWORKS, CATIA, Siemens NX
282
305.27
29.88%… See the full description on the dataset page: https://huggingface.co/datasets/Naitlerand/cad-1000-hours.cad-sft
CAD-SFT
Two subsets for supervised fine-tuning of CAD code-generation models.
Contents
cad-recode-20k (HF dataset config — loadable directly)
20,000 samples randomly drawn from filapro/cad-recode-v1.5 train split. Each CadQuery program is executed with cadquery, the resulting mesh is rendered into a single 4-view PNG, and both the code and the render are stored as a HuggingFace Dataset.
column
dtype
description
stem
string
batch_XX/N (upstream… See the full description on the dataset page: https://huggingface.co/datasets/Hula0401/cad-sft.Zero-To-CAD-1m
Zero-to-CAD 1M
One million executable, interpretable CAD construction sequences synthesized entirely without real-world data.
Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data
Mohammadmehdi Ataei, Farzaneh Askari, Kamal Rahimi Malekshan, Pradeep Kumar Jayaraman
Autodesk Research
Related Resources
Resource
Link
📄 Paper
Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale… See the full description on the dataset page: https://huggingface.co/datasets/0x3/Zero-To-CAD-1m.CAD_tmp_resultscadquarry
CadQuarry
Procedurally generated, execution-validated, fully parametric CadQuery
programs and the geometry they produce. Every part is a pure Python function of
typed, range-bounded parameters and is reproducible bit-for-bit from a seed.
Generator (canonical source): https://github.com/jacobjennings/CadQuarry
generator version v0.6.0, config configs/default.toml
Code license: Apache-2.0 · Data license: CC0-1.0
Each corpus in this dataset is a convenience artifact: the… See the full description on the dataset page: https://huggingface.co/datasets/jacobjennings/cadquarry.