please document more precise the models used in the unit tests

[behrica]

I try to re-run the unit tests, and it is clear that success depends on "precisely" the model which you have used.
So please point even more precise to the "model" (the link on hugging face, for example):

This line

 sess (session env "data/gemma-3-1b-it-ONNX-GQA/onnx/model.onnx" opt)

might not be precise enough, as there could be "various models" having the same name: "gemma-3-1b-it-ONNX-GQA" on huggingface.

Ideally we should have in the tests a full hugging face model name and the "commit-id" as reference, for example:
onnx-community/gemma-3-1b-it-ONNX-GQA : fba787fd660d87319472b5e733ca3e64a38075cf

I believe that in this very early stage of diamond-onnxrt, the code might be sensitive even on minimal changes in the model, and I want to exclude any possibility that I run with even a slidely different version of the model.

[behrica]

I just guessed a bit, and indeed the current test code fails with "latest model": (11 days old)
onnx-community/gemma-3-1b-it-ONNX-GQA

but works with bit older version:

onnx-community/gemma-3-1b-it-ONNX-GQA @ c0afc74bb603f47885fd793c76a73442a3222eec

[blueberry]

Thanks. Will improve that, but the tests are there to make sure that the diamond-onnx works as intended and that we understand onnx specification and make it available from Clojure in a correct way. They are not there to test models and the way that the models work as we think that they do. A particular model can be correct, or completely broken, but that's the issue of the model and our usage of the model, not the issue of diamond-onnx.

[behrica]

I agree, but as we rightly do no commit the large models in git, we can try to be more precise into documenting the exact version in the test , if we know it, so that others can rerun the tests. (even only if to verify there local setup)
ideally we would use a download tool.

[blueberry]

I don't mind a PR related to that :)