In the pre-packaged darknet-nvdla repo that you distribute in the get-darknet.sh script, there are three NVDLA loadable FlatBuffers, yolov3-subgraph0.nvdla, yolov3-subgraph1.nvdla, and yolov3-subgraph2.nvdla. These are generated for the nv_large config used in this project, but I'd love to regenerate these myself for other NVDLA configurations, such as nv_small. Did you manually convert yolov3.weights into a caffemodel, break it into the subgraphs supported by the NVDLA, and then run those 3 caffemodel files through the NVDLA compiler? I'm not entirely sure if that's the right strategy, and it would be great to get some advice before delving into that problem. Thanks!