{"success":true,"data":{"author":"Autodesk Research (Ataei, Askari, Rahimi Malekshan, Jayaraman)","bucketAccessType":"public","bucketName":"ADSKAILab/Zero-To-CAD-1m","bucketPrefix":"","category":"Code & Software Engineering","cloudProvider":"huggingface","createdAt":"2026-11-21T22:15:33.197Z","description":"Zero-to-CAD is a dataset of 999,633 executable, human-readable computer-aided design (CAD) programs for mechanical parts from Autodesk Research, synthesized without any real-world CAD data. An LLM agent worked inside a CadQuery execution environment with tools to execute and validate geometry and to look up the CadQuery documentation, repairing its code over several turns until topology, contiguity, complexity and export checks passed; part descriptions were first generated across 65 categories such as brackets, gusset plates, pulleys, flywheels, housings, covers, clamps and spacers. Unlike sketch-and-extrude datasets such as DeepCAD and Fusion 360 Gallery, the programs use booleans, fillets, chamfers, shells, offsets, lofts, sweeps, revolves, patterns, holes, threads and ribs. Each row has the CadQuery Python source, the list and count of CAD operations, the B-Rep face count, eight 256x256 rendered views, and the exported STL mesh and STEP solid. Splits are 979,633 train, 10,000 validation and 10,000 test rows in 2,241 Parquet files, plus embeddings; about 349 GB (measured 2026-11-21). Stream with Hugging Face datasets. Use it for text-to-CAD, image-to-CAD and CAD code generation. Released under Apache 2.0; cite Ataei et al., Zero-to-CAD, arXiv:2604.24479.","displayTitle":"Zero-to-CAD 1M: One Million Executable CadQuery Parametric CAD Construction Sequences with STEP, STL and Renders (Autodesk Research)","huggingFaceUrl":"https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m","id":"c5d4a72e-7b54-51b6-b9c4-233fa4cbaab2","license":"Apache-2.0","licenseEvidenceUrl":"https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m","objectCountEstimate":2251,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2604.24479","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"ADSKAILab/Zero-To-CAD-1m\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"ADSKAILab/Zero-To-CAD-1m\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=ADSKAILab, license:apache-2.0, size_categories:100K<n<1M, format:parquet, downloads=28422"],"format":"mixed","generatedAt":"2026-11-21T22:31:33.221Z","layout":{"example":"data/test/75_328e3a76afcb4141ae537552f6ce339a_000000_000000-0.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/test/75_328e3a76afcb4141ae537552f6ce339a_000000_000000-0.parquet","truncatedAt":512}],"sampleObjects":["assets/agentic.png","assets/logo.png","assets/samples.png","data/test/75_328e3a76afcb4141ae537552f6ce339a_000000_000000-0.parquet","data/test/75_328e3a76afcb4141ae537552f6ce339a_000001_000000-0.parquet"],"sidecars":[{"bytes":"---\nlicense: apache-2.0\ntask_categories:\n  - text-to-3d\n  - image-to-3d\ntags:\n  - CAD\n  - CadQuery\n  - synthetic-data\n  - construction-sequence\n  - parametric-CAD\n  - 3D-generation\n  - agentic-AI\n  - code-generation\npretty_name: \"Zero-to-CAD 1M\"\nsize_categories:\n  - 1M<n<10M\nlanguage:\n  - en\n  - code\nconfigs:\n  - config_name: default\n    data_files:\n      - split: train\n        path: \"data/train/*.parquet\"\n      - split: validation\n        path: \"data/val/*.parquet\"\n      - split: test\n        path: \"data/test/*.parquet\"\n---\n\n<p align=\"center\">\n  <img src=\"assets/logo.png\" alt=\"Zero-to-CAD\" width=\"100%\"/>\n</p>\n\n# Zero-to-CAD 1M\n\n**One million executable, interpretable CAD construction sequences synthesized entirely without real-world data.**\n\n<p align=\"center\">\n  <img src=\"assets/agentic.png\" alt=\"Zero-to-CAD agentic synthesis pipeline\" width=\"800\"/>\n</p>\n\n> **Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data**\n>\n> [Mohammadmehdi Ataei](https://orcid.org/0000-0002-3399-9696), [Farzaneh Askari](https://orcid.org/0000-0003-0684-1102), [Kamal Rahimi Malekshan](https://orcid.org/0009-0004-1192-4724), [Pradeep Kumar Jayaraman](https://orcid.org/0000-0001-6314-6136)\n>\n> Autodesk Research\n\n## Related Resources\n\n| Resource | Link |\n|----------|------|\n| 📄 **Paper** | [Zero-to-CAD: Agentic Synthesis of Interpretable CAD Programs at Million-Scale Without Real Data](https://arxiv.org/abs/2604.24479) |\n| 📦 **Zero-to-CAD 1M** (this dataset) | You are here |\n| 📦 **Zero-to-CAD 100K** (curated subset) | [ADSKAILab/Zero-To-CAD-100k](https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-100k) |\n| 🤖 **Fine-tuned Model** (Qwen3-VL-2B) | [ADSKAILab/Zero-To-CAD-Qwen3-VL-2B](https://huggingface.co/ADSKAILab/Zero-To-CAD-Qwen3-VL-2B) |\n| 🗂️ **Collection** | [ADSKAILab/Zero-To-CAD](https://huggingface.co/collections/ADSKAILab/zero-to-cad) |\n\n## Overview\n\nZero-to-CAD is a large-scale dataset of **1,000,000 executable CAD construction sequences** generated by an LLM operating inside a feedback-driven CAD environment. Unlike prior datasets limited to sketch-and-extrude workflows, Zero-to-CAD covers a **broad vocabulary of CAD operations** including booleans, fillets, chamfers, shells, lofts, sweeps, and patterns.\n\nEach sample is:\n- ✅ **Executable** — valid CadQuery Python code that runs and produces geometry\n- ✅ **Interpretable** — named parameters, logical construction order, human-readable code\n- ✅ **Geometrically validated** — multi-stage validation (topology, contiguity, complexity, export)\n\n### How It Was Made\n\nWe frame CAD synthesis as an **agentic search problem**: an LLM is placed in a loop with a CadQuery execution environment and equipped with three tools:\n\n1. **`execute_and_validate`** — runs code, checks geometric validity, returns structured feedback\n2. **`lookup_documentation`** — TF-IDF retrieval over CadQuery API docs\n3. **`grep_documentation`** — regex-based precise syntax lookup\n\nThe agent iteratively generates, executes, and repairs code across multiple turns until validation succeeds. A two-stage protocol first generates diverse part descriptions (65 categories), then synthesizes executable code for each.\n\n<p align=\"center\">\n  <img src=\"assets/samples.png\" alt=\"Sample CAD models from Zero-to-CAD\" width=\"100%\"/>\n</p>\n\n## Dataset Details\n\n### Generation Statistics\n\n| Metric | Value |\n|--------|-------|\n| Total sequences | 999,633 |\n| Total tokens processed | 60.2B |\n| Generated tokens per design (mean / median) | 5,638 / 5,089 |\n| Function calls per conversation (mean) | 4.34 |\n| Zero-shot success rate | 22.3% |\n| Attempts before success (mean / median) | 3.30 / 3 |\n| Minimum B-Rep face count | 7 |\n\n### Splits\n\n| Split | Samples |\n|-------|---------|\n| Train | 979,633 |\n| Validation | 10,000 |\n| Test | 10,000 |\n\n### Data Fields\n\nEach sample contains:\n\n| Field | Type | Description |\n|-------|------|-------------|\n| `uuid` | `string` | Unique identifier (deterministic, derived from content hash) |\n| `cadquery_file` | `string` | Executable CadQuery Python","encoding":"utf8","key":"README.md","name":"README.md","size":11102,"truncated":true}],"sourceObjectKey":"data/test/75_328e3a76afcb4141ae537552f6ce339a_000000_000000-0.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"ADSKAILab/Zero-To-CAD-1m\")\nprint(ds)","starterSamplePath":"assets/agentic.png","totalObjects":2251},"summary":"Zero-to-CAD is a dataset of 999,633 executable, human-readable computer-aided design (CAD) programs for mechanical parts from Autodesk Research, synthesized without any real-world…","title":"Zero-to-CAD 1M: One Million Executable CadQuery Parametric CAD Construction Sequences with STEP, STL and Renders (Autodesk Research)","updatedAt":"2026-11-22T00:50:23.123Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/ADSKAILab/Zero-To-CAD-1m"}}