Pyre Code Ml Practice

If you want to actually understand how modern ML systems work under the hood, this gives you 68 coding problems that make you implement the internals yourself. You'll build pieces of Transformers, diffusion models, RLHF algorithms, and inference optimizations like Flash Attention and KV caching. It runs entirely in your browser with a local FastAPI grading service, no GPU needed. The setup is straightforward (one shell script or Docker Compose), and you get instant pass/fail feedback on each submission. It's self-hosted, so you can add your own problems or modify the test cases. Honestly, this is what LeetCode would look like if it cared about ML engineering instead of inverting binary trees.