ml-training-skill

PyTorch training recipes for all neural network domains. Use when training or fine-tuning models, debugging loss spikes or NaN gradients, resolving OOM errors, choosing model architectures, selecting optimizers, tuning learning rates, or optimizing GPU throughput. Covers LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, and genomics. Do NOT use for pandas data analysis, API serving, unit testing, or scRNA-seq differential expression.