Model loader architecture and implementation skill for local AI apps, especially AIWF. Use when coding or reviewing model loading, precision and dtype policy, quantization selection, GGUF/safetensors/Diffusers/Transformers/Nunchaku/vLLM/llama.cpp/TensorRT backend choice, LoRA adapter compatibility, VAE/text-encoder precision, model manifests, preflight checks, loader fallback rules, and runtime tests.
