NVIDIA Releases TensorRT Model Connect in Public Preview: Hugging Face Checkpoint to Native C++ Inference in Two Commands

NVIDIA launched TensorRT Model Connect in public preview, enabling direct conversion of Hugging Face or local checkpoints to TensorRT inference with two commands, bypassing ONNX. The open-source tool produces a .bundle artifact for native C++ inference without PyTorch dependency, targeting Linux aarch64 currently.