Rudra-narayan-muduli-001/One_Use ? reverse-engineered prompt

Build me a single Python package that can open and inspect machine learning model files in GGUF, Safetensors, ONNX, and PyTorch formats from one simple API.

I want it to automatically figure out the file type by reading the file itself, not by trusting the extension, then load it in the most memory efficient way possible, ideally with zero copy behavior when it can. It should let me quickly see model info, list tensors and shapes, and then either run normal inference or generate text for GGUF models. Please make it feel lightweight to import, and only bring in the heavy dependencies when they are actually needed.

It should also work with local model files, support common hardware options when available, and have a clean structure so it is easy to extend later. If you need to check current docs for any of the model libraries, go ahead and look them up online.