training-nnlisted

# Training Neural Networks Neural net training fails silently: the code runs, throws no exceptions, and still trains something wrong. Use this skill to work through a disciplined recipe instead of guessing at hyperparameters. ## Quick Checklist First Before anything else, rule out the most common silent bugs — see [most-common-neural-net-mistakes.md](most-common-neural-net-mistakes.md): - Didn't try to overfit a single batch first - Forgot to toggle train/eval mode - Forgot `.zero_grad()` before `.backward()` - Passed softmaxed outputs to a loss that expects raw logits - Wrong `bias=False`/`True` interaction with BatchNorm - Confusing `view()`/`permute()` ## Full Recipe For the complete step-by-step process — becoming one with the data, building an end-to-end skeleton with dumb baselines, overfitting, regularizing, tuning, and squeezing out final gains — see [recipe-for-training-nn.md](recipe-for-training-nn.md). Use the recipe's step structure (1. Become one with the data → 2. End-to-end skeleton + baselines → 3. Overfit → 4. Regularize → 5. Tune → 6. Squeeze out the juice) as the default order of operations for any new training pipeline.