Do these models run out-of-the-box on the silicon, including pre/post processing? 

Yes — every model in the garden comes with a pre-validated runtime package, including compatible ONNX/quantized models, pre-processing pipelines (resize, normalization, tokenization), and post-processing (NMS, depth scaling, keypoint decoding, etc.).