Launching today
GitHub
Build AI the way you think—Python library for AI development
4 followers
Build AI the way you think—Python library for AI development
4 followers
Omage is an open-source Python library that makes building, training, and deploying AI models simple and readable. Instead of 50+ lines of PyTorch boilerplate, you write clean expressive code. Omage handles the rest. Key features: • Model Zoo — ResNet, BERT, GPT-2, YOLOv8 ready to use • AI Loops — polish, seek, grow, flow • Progress bars + checkpointing + early stopping • GPU auto-detection • 61 tests on real datasets Currently in Alpha (v0.1.1) — core API is stable.
The clean API looks great, and the Model Zoo covering everything from ResNet to YOLOv8 is a solid starting point. One thing that would really help adoption is built-in support for mixed precision training out of the box, maybe just a flag in the AI Loops config. Most users expect it now, and wiring it manually around your loop definitions is the first thing I'd have to do for any real training run.
Tried Omage over the weekend and was surprised how the polish/seek/grow/flow loop just works without me writing training boilerplate. The YOLOv8 example trained cleanly on my laptop and the progress bars plus early stopping saved me a headache. Nice work.
The "AI Loops" naming (polish, seek, grow, flow) is such a nice touch, way more memorable than generic fit/train/eval jargon. Love that the core API stayed stable through alpha.