SOFIE Developments - Inference Code Generation for Deep Learning models
- Mentors
- Neel Shah, Lorenzo Moneta, Sanjiban Sengupta
- Organization
- CERN-HSF
- Technologies
- python, c++, pytorch, onnx, ROOT
- Topics
- machine learning, deep learning, physics, data analysis, code generation, Fast Inference
SOFIE aims to streamline the integration of deep learning models into the ROOT scientific software framework through efficient inference code generation. By implementing missing ONNX operators, particularly those used in transformer models, we will enhance TMVA's capabilities for building complex neural networks. Our project will deliver modular operator classes adhering to ONNX standards, accompanied by thorough unit testing, enabling seamless deployment of deep learning models in physics software and analysis frameworks.