I may be missing something painfully obvious here... but is there any place or way that you can quickly decipher how you create the input tensor from raw data to feed into a model you're working with?

For example, if I grabbed or was given an onnx model, is there any way to discover it's input tensors or is there a central repository of example input tensors I can check?

Every time I find a new pretrained model it semes like a majority of my time is spent just figuring out the input and output tensors and it rarely seems to be properly documented in a reasonable way even on huggingface.