Michele LaurelliBlog

TensorRT

TensorRT: An NVIDIA SDK that optimizes and runs deep learning models on compatible GPUs. It applies layer fusion, reduced precision and hardware-specific kernels. Build stages and dynamic profiles must be checked against input shapes and runtime version. Examples include TensorRT engine, FP16 inference on NVIDIA GPUs, CNN optimization.

In italiano: TensorRTLegge IV · Loss, gradiente e ottimizzatori

It applies layer fusion, reduced precision and hardware-specific kernels. Build stages and dynamic profiles must be checked against input shapes and runtime version.

Examples

  • TensorRT engine
  • FP16 inference on NVIDIA GPUs
  • CNN optimization