Our great sponsors
-
WorkOS
The modern identity platform for B2B SaaS. The APIs are flexible and easy-to-use, supporting authentication, user identity, and complex enterprise features like SSO and SCIM provisioning.
-
pytorch-lightning
Discontinued Build high-performance AI models with PyTorch Lightning (organized PyTorch). Deploy models with Lightning Apps (organized Python to build end-to-end ML systems). [Moved to: https://github.com/Lightning-AI/lightning] (by PyTorchLightning)
DL is all experimentation. You need a framework that makes modification easy, trackable and reproducible. The closest to achieving that these days is based on pytorch lightning with hydra. Check out this video and repo: https://www.youtube.com/watch?v=w10WrRA-6uI https://github.com/Erlemar/pytorch_tempest
Thanks for sharing this, this is probably the best thing here. What makes Hydra really cool is the config system, which is done using OmegaConf (https://github.com/omry/omegaconf), and I especially enjoy the option of defining the configs using Python Data Classes.
When I look at the pytorch lightning animation, the stuff on the left for me is easy to follow and the code on the right formatted into classes is hard. My goal is to to start thinking and coding more like the code on the right. What I typically find hard with reading through code where everything is inside classes, methods, functions, decorators etc (i.e. the code on the right) is that there will be a place that executes all these methods in a linear way, but I keep having to scroll up to the class to see what it is actually doing. On the left I can just read through the code top to bottom. I even find myself copying the code out of classes the first time I read it so it executes like the code on the left :P I feel like what I'm doing is the equivalent of typing with only my index fingers…