PyTorch Neural Network wrapper to help transition existing projects into the Mojo programming language. https://github.com/mytechnotalent/mojonet
#mojo
#mojonet
final bolt
austere pawn
Interesting. So it's basically a Pytorch wrapper?
...
var torch = Python.import_module("torch")
...
Have you done any benchmark tests, comparing it to using Pytorch directly?
By the way, there is https://discord.com/channels/1087530497313357884/1238547362012729355 pure Mojo 😉
final bolt
Oh thank you. I have not I wanted to contribute what I could to get those who use PyTorch and easy way to integrate their existing projects in Mojo.
final bolt
@austere pawn I would love if you had the time to do a PR for the next release to add in benchmarking functionality and unittests if you had an opportunity being you are more senior in Mojo as this would help advance exsiting AI Devs who work with PyTorch to consider the Mojo platform.
high dawn
Isn't Mojo's bindings should be to C++ libtorch library? for the speed..
final bolt
@high dawn the point of this repo is for AI Developers who are comfortable with PyTorch to have a quick framework to get their project working with the Mojo platform. The point of this repo is wider adoption to Mojo as there are a large number of AI Developers I have introduced to Mojo and they are extremely hesitant without some helper framework to get them started and that is what I am trying to do to advance the Mojo ecosystem.
tribal graniteBOT
Congrats @final bolt, you just advanced to level 1!
austere pawn
<@989606857180057652> I would love if you had the time to do a PR for the next r...
Sorry, but I even haven't had the chance to set this up for my own projects. I've been juggling too many projects at once, and don't have the expertise yet to establish meaningful benchmark tests yet.
final bolt
No problem @austere pawn appreciate all the work you all do!
final bolt
I attempted to write unittests however I was unable to import my mojo package properly from the tests folder. I tried to follow other examples but was unsuccessful.
final bolt
I was successful in creating some unittest @cold dock and will push them up tomorrow. I am hoping the larger Python Community who uses PyTorch is excited to try this out.
final bolt
Attempted to use mps however it crashes. I created a detailed bug report as I am trying to get this PyTorch wrapper to work with cuda and mps to make it more powerful. https://github.com/modularml/mojo/issues/3092
final bolt
@austere pawn I am trying to use mps which is MAC's cuda alternative in my PyTorch wrapper. Are these capabailities not implemented yet in Mojo? I created a bug for this reason and wanted to better understand. Thank you for your time!