Merge pull request #974 from anarkiwi/doc2

Orin docs.

docker/Dockerfile.orin-torchserve

@@ -7,3 +7,14 @@ RUN /usr/local/bin/torchserve --help
 COPY torchserve/config.properties /torchserve/config.properties
 COPY torchserve/torchserve-entrypoint.sh /torchserve/torchserve-entrypoint.sh
 CMD ["/torchserve/torchserve-entrypoint.sh"]
+
+# 0. Always start from a clean sdkmanager, installation of Ubuntu 20.04. DO NOT attempt an in-place JetPack upgrade, DO NOT try to install docker-ce, and DO NOT try to re-install/replace CUDA without re-imaging the entire system.
+# 1. Install JetPack via apt: https://docs.nvidia.com/jetson/jetpack/install-jetpack/index.html#install-jetpack. apt show nvidia-jetpack should show 5.1.2.
+# 2. sudo apt-get disc-upgrade, reboot
+# 3. git clone https://github.com/iqtlabs/gamutrf
+# 4. cd gamutrf ; vi tests/test_torchserve.sh
+# Change the docker run command line as follows:
+# a) Change gamutrf-torchserve to gamutrf-orin-torchserve
+# b) Insert —-runtime nvidia as first args, after docker run
+# 5. sudo apt-get install python3-pip
+# 6. Run tests/test_torchserve.sh. A successful test will print JSON prediction results. It may take 60s after the container starts, for it to return inference results as the model server installs ultralytics