ONNX
ONNX
Also known as: Open Neural Network Exchange
ONNX (Open Neural Network Exchange) is the open format that lets a model trained in PyTorch run in TensorFlow, OpenVINO, TensorRT, or any ONNX Runtime — without vendor lock-in. Operator coverage spans CNNs, transformers, RNNs, and graph operations. ONNX Runtime ships its own cross-platform inference engine (CPU, CUDA, DirectML, CoreML,…
Definition
ONNX (Open Neural Network Exchange) is the open format that lets a model trained in PyTorch run in TensorFlow, OpenVINO, TensorRT, or any ONNX Runtime — without vendor lock-in. Operator coverage spans CNNs, transformers, RNNs, and graph operations. ONNX Runtime ships its own cross-platform inference engine (CPU, CUDA, DirectML, CoreML, WebGPU). FI Tech treats ONNX as the canonical handoff format between the data-science team in Riyadh and the field deployment team — every model leaves the lab as ONNX, then gets compiled to TensorRT, OpenVINO IR, or Hailo HEF for the target box.
