Hi, I’m really new to this forum (hence I don’t know where to start a new thread) and also new to NVIDIA developer, ML and Modulus, I actually haven’t used it yet and that’s why I have the following question:
can I install modulus on a Jetson Xavier developer card or a jetson nano? as those are hardware already present in my project and my supervisors want me to use them.
Most of Modulus will function is PyTorch can function using the bare-metal install. Only a few features will not be available. However, the examples provided may need to have their parameters or model size adjusted based on the hardware you are using.
Next time, please create a new post by clicking on the relevant category (e.g. technical support) then clicking “New Topic”. Thanks.
Thanks for your reply. I can run and debug the code in /examples folder by loading the modulus image first into the docker container. I want to add new pdes in modulus.eq.pdes now so that I can run test new examples.
In order to develop the source code of Modulus, I need to mount the source code in the docker container by using the -v option. Do I need to load the modulus image into the docker first? If so, when I modify the source code of Modulus, the example script will import the modulus module in the source code or the modulus module in the modulus image?
I wish that the Modulus team can provide more detailed documentation on the developing procedure for the Modulus so that more users can use and contribute to the Modulus package.
Thanks for the feedback, we are actively working on making it easier for people to contribute / customize / extend Modulus.
Regarding you question, individual responses are below:
In order to develop the source code of Modulus, I need to mount the source code in the docker container by using the -v option.
Correct, -v mounts a directory on your local machine into the docker image. This allows you to edit files on in your file system as opposed to inside the image itself. Any edits to these files outside the docker container will be updated inside the container.
Do I need to load the modulus image into the docker first?
Yes, you want to use the Modulus container because this has all the dependencies installed. The idea is to mount your own Modulus source code, then overload the container Modulus via a python install.
If so, when I modify the source code of Modulus, the example script will import the modulus module in the source code or the modulus module in the modulus image?
Correct, if you python setup.py develop using your mounted version inside the container this will be prioritized when running the examples (when you import modulus). The “develop” install will allow you to make edits to the source code and immediately have the changes impact and script that imports Modulus.
Hopefully this helps, let me know if you have additional questions.