Fork me on GitHub
#data-science
<
2021-07-21
>
lepistane16:07:06

Hello, Uber beginner here. I am not sure if this is the best place to ask but here it goes... My friend created some AI/Computer Vision models as a joke and he would like to deploy them and be accessible outside of his computer. I thought this would be fun opportunity for me to do that. I am thinking about making them accessible via API. His project is done in Python and py torch i believe. So my question is are models language agnostic? Can i setup Clojure webapp that just serves those models or i need to use https://github.com/clj-python/libpython-clj for example to do that? Would it just be easier if i create this webapp in Python instead of Clojure? Are there any resources/guides for this being done in Clojure can it be done at all?

respatialized16:07:33

There are formats that allow you to export from PyTorch so that a trained model can be run in a given alternative runtime: https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html If you were to go down such a route then you could use the Java API for the ONNX runtime: https://www.onnxruntime.ai/docs/reference/api/java-api.html But overall I think the answer to your second to last question might still be "yes", because it adds fewer moving parts to your stack - but if it's just a hobby project it could be a good opportunity to get familiar with the Clojure DS/ML/DL ecosystem