Torchcode Pytorch Interview Practice

If you're prepping for ML engineering interviews or just want to sharpen your PyTorch fundamentals, this gives you 40 LeetCode-style problems focused on implementing everything from ReLU and softmax to multi-head attention and GPT-2 components from scratch. The torch_judge package auto-grades your implementations with correctness checks, gradient verification, and timing feedback. You can run it in Colab, Hugging Face Spaces, or locally via Docker. The problems range from numerical stability tricks in softmax to full transformer blocks, and each notebook includes templates plus reference solutions. It's honestly the kind of drilling that separates people who can use PyTorch from people who understand what's happening under the hood.