Torch Clinic — PyTorch training-code review

Paste your PyTorch model and training code and get a deep-learning engineer's review: a train-ready / hardening-recommended / silent-failure-risk posture, the module and loop inventory, prioritized findings with corrected Python, quick wins and the focus areas to work through first. A free instant browser prescan flags missing zero_grad, backward without step, softmax feeding CrossEntropyLoss, loss accumulated with the autograd graph attached, validation never switched out of train mode, missing seeds, hardcoded CUDA, starved DataLoaders and missing checkpoints before you spend anything. Derived from the @pytorch/pytorch skill (BSD-style license).

Every public app is built from a security-scanned skill and must pass a clean scan — skill and frontend — before it can be listed. Have a skill of your own? Turn it into an app — or read the step-by-step walkthrough.