############ HOOPS AI 1.1 ############ **Release 1.1 (June 15, 2026)** What's New in HOOPS AI 1.1.0 ============================== ---- HOOPS AI 1.1 introduces the **SIGNAL architecture** for HOOPS Embeddings, designed to deliver more robust geometric representation, improved shape matching, and higher-quality CAD similarity results. ---- * **Python 3.12 and PyTorch 2.9** - HOOPS AI now targets Python 3.12 and PyTorch 2.9. This brings support for the latest NVIDIA GPU architectures (Ada Lovelace, Blackwell) and improved performance across the board. * **Migration from DGL to PyTorch Geometric** - DGL (Deep Graph Library) has been removed in favour of `PyTorch Geometric `__. DGL is no longer actively maintained for the use cases covered by HOOPS AI. PyTorch Geometric is the actively developed standard for graph neural networks in the PyTorch ecosystem. * **Context Layer** - Introduced the Context Layer, a new capability for inferring engineering context from nearest CADSearch neighbors. Supports prediction of metadata such as Material, Manufacturing, Label, Cost, Weight, Lead Time, and other user-defined attributes. Machine Learning and Retrieval ================================ HOOPS Embeddings SIGNAL Architecture ------------------------------- * Preview of the **SIGNAL architecture** for HOOPS Embeddings. * Enhanced retrieval quality based on internal benchmark results. * Added a second-stage reranking step to improve the relevance of top search results. * The reranker refines the order of the top candidates to improve the accuracy and relevance of CADSearch results for a given query. ---- Context Layer (Preview) ============= * Introduced the Context Layer, a new capability for inferring engineering context from nearest CADSearch neighbors. * Supports prediction of metadata such as Material, Manufacturing, Label, Cost, Weight, Lead Time, and other user-defined attributes. * Consumes ``VectorHit`` results from CADSearch and retrieves neighbor metadata through a user-supplied ``ContextProvider`` layer. * Added ``ContextPrediction`` objects for each requested context key. Apply a status policy to deliver ready-to-propose, needs-review, insufficient-evidence. ---- Breaking Changes ================ Python and PyTorch ------------------- HOOPS AI 1.1.0 requires **Python 3.12** and **PyTorch 2.9**. Earlier Python versions (3.11 and below) are not supported. DGL to PyTorch Geometric ------------------------- DGL is no longer a dependency. If you have custom models or data loaders that use DGL graph objects (``dgl.DGLGraph``), they must be migrated to PyTorch Geometric (``torch_geometric.data.Data``). The built-in HOOPS AI models and pipeline components have been updated and require no changes on your side if you are using the standard flow. ---- Environment Setup ================= The installation workflow has changed. Conda is no longer required. HOOPS AI 1.1.0 uses a standard Python virtual environment. See :doc:`/getting_started/evaluate` for full installation instructions. ---- Supported Platforms =================== See :doc:`/getting_started/supported_platforms` for the full platform and driver requirements.