Question regarding precision

[FruitClover]

Hi, for the readme example:

PyTorch prediction
[('Labrador retriever', 70.65674591064453), ('golden retriever', 4.988346099853516), ('Saluki, gazelle hound', 4.477451324462891)]
torch-mlir prediction
[('Labrador retriever', 70.6567153930664), ('golden retriever', 4.988325119018555), ('Saluki, gazelle hound', 4.477458477020264)]

are there are any notes somewhere for the precision difference, maybe regarding high-level stages (e.g. pytorch --> torch mlir, torch mlir --> linalg/stablehlo/tosa backend, backend output --> LLVM IR and so on), concreate operations or CPU/GPU execution? (not just for the resnet model, this is just an example output)