nimtorch
nimtorch copied to clipboard
sinkingsugar
Is this project still active
Hi,
I think a working binding to pytorch from nim could be very valuable to support the use of nim in data science. This project seems to be inactive for a long time now and it is not using the current version of pytorch - any chance that it will be updated?
Hi, I'm definitely active but not active in this project indeed. Not using it personally for now and no sponsorships or particular interest around.
Maybe this is the right time to revive it? The main issue I had was the installation of the torch libraries but I think this could be much simpler now with the native libraries available as downloads from pytorch so a complete nimble package would be possible?
I'm working as a data scientist in a big company and I think that nim would be a great option if it had something like nimtorch as a kind of nucleus around which other tools can be build.
Fast Tensor operations from pytorch could be exactly this but the install needs to be seamless.
I will try to advertise the project a bit and if you have a patreon account or something similar I will do my best to help sponsor your work. I am not very good with nim (the only thing I tried so far was a wrapper for Arrayfire) but I would also try to contribute here if you like.
You could see if Arraymancer fits your needs: https://github.com/mratsim/Arraymancer
Or https://github.com/SciNim/minitorch might be a better way to move forward with pytorch bindings.
EDIT: #science channel on the Nim Discord or the IRC channel could also be a great to ask questions regarding Nim and data science.
Maybe this is the right time to revive it? The main issue I had was the installation of the torch libraries but I think this could be much simpler now with the native libraries available as downloads from pytorch so a complete nimble package would be possible?
The whole point of nimtorch was not just being "bindings", it was using just the core of pytorch (ATEN back then, now probably changed name), and we re-implemented from scratch in nim the boilerplate.
It was ambitious, today I would just use the C++ layer of pytorch and call it a day.
So yeah maybe minitorch which is just bindings might be ok.. but good luck binding to c++ with nim really, it's a painful experience 😄
Anyway contributions are always welcome.
Thanks for the update. I have looked at minitorch which seems to be very raw at the moment but let's see where it goes. Anyhow, many thanks for this inspiring work!