Today's AI Frameworks & SDKs: Fastest-Growing Projects — October 02, 2026
Today's AI Frameworks & SDKs space on GitHub continues to showcase a diverse range of innovative projects that cater to various aspects of machine learning and artificial intelligence. From high-performance communication libraries for specific hardware accelerators to frameworks that enable interactive cinematic experiences with AI-generated assets, the ecosystem is thriving with unique solutions addressing different challenges in the field.
The deepseek-ai/DeepEP-Ascend project stands out this week with a growth score of 65.50 and 199 stars. It offers a high-performance communication library designed for machine learning training and inference on Huawei Ascend NPUs, making it particularly appealing to developers looking to leverage the unique capabilities of these hardware accelerators.
Blendi-remade's dioramas project has garnered significant attention with a growth score of 60.00 and 180 stars. This open-source framework allows users to create cinematic, interactive 3D websites featuring AI-generated assets, specifically utilizing Nano Banana 2 and Meshy 7.1 on fal, providing an engaging visual experience for website visitors.
The wfzyx/von repository has seen notable activity with a growth score of 56.82 and 805 stars. It is the open-source System One decision model that offers sub-15ms, non-autoregressive performance as a local drop-in alternative to TypeSafe Jev, making it an attractive solution for developers looking to optimize their decision-making processes.
PSRben's VisionHOPE project has attracted 409 stars and maintains a growth score of 52.90. This official PyTorch implementation of VisionHOPE focuses on visual backbones as self-modifying learning systems, offering researchers and practitioners an advanced framework for developing adaptable neural network architectures.
AbdelStark's awesome-typesafe-jev repository has amassed 554 stars and a growth score of 52.47. It serves as a comprehensive field guide to TypeSafe’s System One model, providing SDKs, live demos, agent tools, and independent evaluations to help users understand and implement the model effectively.
maanHimself's OpenDLSS-NR project has gained traction with 740 stars and a growth score of 39.46. This Vulkan reimplementation offers a bit-exact replication of NVIDIA’s DLSS 5 Neural Rendering network, providing developers with an open-source alternative to explore and customize.
CaptureGrubEnchant's SolidWorks MCP Server project has seen significant development activity with 182 stars and a growth score of 34.43. It connects an AI assistant to a running SolidWorks instance, enabling users to perform tasks such as sketching, extruding, filleting, exporting STEP/STL files, and generating macros directly within the software.
MiniMax-AI's OpenAgentCore project has attracted interest with 113 stars and a growth score of 33.36. This open-source implementation of the OpenAI Agents API offers multiple native harnesses for self-hosted use, providing developers with flexibility in integrating AI agents into their applications.
yifanzhang-pro's recurrent-looped-tranformer repository has gained substantial popularity with 905 stars and a growth score of 33.20. It is the official project page for Recurrent Looped Transformer (RLT), offering researchers and developers access to an advanced transformer model designed for specific tasks.
Continuum-AI-Corp's OrcaBonsai-27B-Uncensored repository has seen steady interest with 562 stars and a growth score of 31.21. It focuses on runtime behavioral ablation for compressed LLMs, specifically targeting the Ternary Bonsai 2 27B model without modifying or re-quantizing weights, offering valuable insights into optimizing large language models.
These projects highlight the dynamic nature of AI Frameworks & SDKs, catering to a wide range of applications and audiences from hardware-specific optimizations to interactive web experiences and advanced modeling techniques.
The deepseek-ai/DeepEP-Ascend project stands out this week with a growth score of 65.50 and 199 stars. It offers a high-performance communication library designed for machine learning training and inference on Huawei Ascend NPUs, making it particularly appealing to developers looking to leverage the unique capabilities of these hardware accelerators.
Blendi-remade's dioramas project has garnered significant attention with a growth score of 60.00 and 180 stars. This open-source framework allows users to create cinematic, interactive 3D websites featuring AI-generated assets, specifically utilizing Nano Banana 2 and Meshy 7.1 on fal, providing an engaging visual experience for website visitors.
The wfzyx/von repository has seen notable activity with a growth score of 56.82 and 805 stars. It is the open-source System One decision model that offers sub-15ms, non-autoregressive performance as a local drop-in alternative to TypeSafe Jev, making it an attractive solution for developers looking to optimize their decision-making processes.
PSRben's VisionHOPE project has attracted 409 stars and maintains a growth score of 52.90. This official PyTorch implementation of VisionHOPE focuses on visual backbones as self-modifying learning systems, offering researchers and practitioners an advanced framework for developing adaptable neural network architectures.
AbdelStark's awesome-typesafe-jev repository has amassed 554 stars and a growth score of 52.47. It serves as a comprehensive field guide to TypeSafe’s System One model, providing SDKs, live demos, agent tools, and independent evaluations to help users understand and implement the model effectively.
maanHimself's OpenDLSS-NR project has gained traction with 740 stars and a growth score of 39.46. This Vulkan reimplementation offers a bit-exact replication of NVIDIA’s DLSS 5 Neural Rendering network, providing developers with an open-source alternative to explore and customize.
CaptureGrubEnchant's SolidWorks MCP Server project has seen significant development activity with 182 stars and a growth score of 34.43. It connects an AI assistant to a running SolidWorks instance, enabling users to perform tasks such as sketching, extruding, filleting, exporting STEP/STL files, and generating macros directly within the software.
MiniMax-AI's OpenAgentCore project has attracted interest with 113 stars and a growth score of 33.36. This open-source implementation of the OpenAI Agents API offers multiple native harnesses for self-hosted use, providing developers with flexibility in integrating AI agents into their applications.
yifanzhang-pro's recurrent-looped-tranformer repository has gained substantial popularity with 905 stars and a growth score of 33.20. It is the official project page for Recurrent Looped Transformer (RLT), offering researchers and developers access to an advanced transformer model designed for specific tasks.
Continuum-AI-Corp's OrcaBonsai-27B-Uncensored repository has seen steady interest with 562 stars and a growth score of 31.21. It focuses on runtime behavioral ablation for compressed LLMs, specifically targeting the Ternary Bonsai 2 27B model without modifying or re-quantizing weights, offering valuable insights into optimizing large language models.
These projects highlight the dynamic nature of AI Frameworks & SDKs, catering to a wide range of applications and audiences from hardware-specific optimizations to interactive web experiences and advanced modeling techniques.