NVIDIA releases cuDLA Python module for Jetson AGX Orin DLA inference

The NVIDIA cuda.bindings.cudla module exposes the cuDLA runtime directly to Python, enabling developers to execute native deep learning inference on the NVIDIA Deep Learning Accelerator on Jetson AGX Orin without writing C++ code. The module supports hybrid-mode workflows where CUDA streams coordinate DLA execution with GPU operations. This guide walks through installation, compiling a model with trtexec, module loading, tensor introspection, memory registration, task submission, and synchronization. Open article →