The Chimera SDK offers a comprehensive suite for developing and optimizing AI applications destined for the Chimera GPNPU. This software development toolkit streamlines the integration of both machine learning and traditional C++ code, providing developers with the flexibility to seamlessly transition applications from conceptualization to execution. Offering support for a wide range of ML models and formats, the Chimera SDK efficiently compiles ML graph code into executable C++ code, ensuring compatibility with existing system infrastructures.
Emphasizing ease-of-use and flexibility, the Chimera SDK lets developers access its powerful tools either online via the DevStudio platform or offline through a standalone installation on private servers or clouds. The environment supports compiling, simulating, and profiling of code, enabling developers to optimize performance while minimizing power usage. Additionally, its integration with the Chimera LLVM C++ Compiler and Instruction Set Simulator provides detailed insights into application performance, helping in fine-tuning code to meet specific design requirements.
The SDK also provides a robust graph compiler that accepts input graphs in ONNX format from leading ML frameworks like TensorFlow and PyTorch. It performs comprehensive optimizations on these graphs, enhancing execution efficiency and reducing resource consumption. With the Chimera SDK, developers have the capability to redefine ML operator use and memory allocation dynamically, offering unprecedented flexibility and control over AI application deployment.
