ML Training Recipes Battle-tested patterns for PyTorch training across domains. Drawn from production codebases (Karpathy's autoresearch/nanochat, torchvision, HuggingFace) and modern training practice. Reference files (read when needed) - — Transformer/LLM architecture code patterns, weight init - — Muon, AdamW hybrid, per-group LR, compiled optimizer steps - — Vision, diffusion, contrastive, distributed, checkpointing, data loading - — Scaling laws, compute budget tables, decision trees, DGX Spark - — Drug discovery, protein models, medical imaging, genomics, clinical NLP - — Autonomous exp…