Hello world of "is this gpu on"; construct a an 8 million element tensor and compute sigmoid on it. Roughly 5x faster on the GPU than CPU. #pytorch

Though I guess I should get a different computer with a #pytorch compatible #GPU. Are the >M1 Macbooks OK for this? If not, what laptops do people recommend for playing around with machine learning?

I think I'm starting to develop a whole vendetta against #PyTorch with the new dependencies in 1.13.0. At first it was, "Hm, this Docker image is really big," and now it's cascaded into all the Docker images that are repeatedly getting built are ginormous and eating up all the disk space, and things that used to just work are failing. :blobfoxangrylaugh:

That's unexpected: Microsoft releases #scala bindings for #pytorch
https://github.com/microsoft/scala_torch

didn't know that MSFT uses scala much, and happy to find out)

Ugh, installing #PyTorch 1.13.1 is cramping my style. I have a custom Docker image with PyTorch installed. Here are image sizes with different torch configurations from largest to smallest:
* torch==1.13.1: 7.21 GB
* torch==1.12.1: 4.52 GB
* torch==1.13.1 (cpu): 3.03 GB
* no torch: 2.09 GB

I can see the new lines in the torch setup.py that bring in huge NVIDIA dependencies, but I've not found a lot of info on why.

Seriously though, why is #PyTorch is great, but why is it so big?!

If you're worried about things like the recent #pytorch #security issue, or generally feel paranoid about where your packages come from, I've broken ground on an #opensource side project that will help with this! Starting with pip (done!) and npm (upcoming!) support initially

If using nightly builds of #Pytorch, take stock of your resources & recycyle secrets.. there was a supply chain attack impacting users of "torchtriton" library.

https://pytorch.org/blog/compromised-nightly-dependency/

#security #vulnerability #securitynews #AI #machinelearning

The developers of PyTorch (a popular machine-learning framework) recently identified a malicious dependency confusion attack on the open-source project. Security teams are advised to check for infected resources and rotate any exposed keys.. https://www.wiz.io/blog/malicious-pytorch-dependency-torchtriton-on-pypi-everything-you-need-to-know #PyTorch #Security

PyTorch: Machine Learning toolkit pwned from Christmas to New Year

https://nakedsecurity.sophos.com/2023/01/01/pytorch-machine-learning-toolkit-pwned-from-christmas-to-new-year/ #Pytorch

they call it #PyTorch because you have to torch your machine afterwards

Yay the upgrade script worked (after needing to comment out the !wget and !gzip lines but not any of the other magic, weird) and now it's uninstalling the old version of #tensorflow... now #pytorch has a weird error: RuntimeError: THPDtypeType.tp_dict == nullptr INTERNAL ASSERT FAILED at "../torch/csrc/Dtype.cpp":135, please report a bug to PyTorch.

If someone here can give me an explanation for "why #mxnet is not more widely used while #Tensorflow and #Pytorch are slower?". I like TF and PT but mxnet is IMHO not used enough.