[docs] model testing

[stevhliu]

refactors the testing docs to be more contributor-facing organized around writing model tests instead of being a collection of pytest usage examples and CI maintenance. also updates the pr checks doc so contributors are better prepared to pass the checks on first try

testing.md

• adds CausalLMModelTest and VLMModelTest paths (should use these more often)
• explains using ModelTester and ModelTest for non-causal and vlm models
• explains what mixins provides
• explains using real pretrained weights versus tiny models
• flags for controlling what to test

---

pr_checks.md

• clear make command flow
• describe what make check-repo does
• mention slow tests
• condense # Copied from section

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[vasqu]

@ydshieh @tarekziade could you do a first pass? Would take a look then, quite a lot to catch up on 😬

[tarekziade]

Great work, thanks a lot! I took a first pass focusing on the CI aspects.

A couple of meta remarks:

It might be better to avoid referring to CircleCI directly in the docs, since where/how we run CI is evolving. Maybe we can use a more generic term like “Hugging Face CI”?
Similarly, the detailed description of CI jobs could become outdated fairly quickly given the changes we’re planning. It might be more useful to emphasize the Makefile targets instead, as we’re aiming to make them the source of truth for both CI and local development.

[tarekziade]

LGTM just one nit about mlinter

[vasqu]

Some initial comments from my side, sorry a bit jumbled thoughts here and there. Hope it's clear enough otherwise happy to clarify

[vasqu]

Should we have a small intro that gives an intro what these subsections are, i.e. referencing the explicit names of our repo checks

[vasqu]

I feel like we are missing a TOC or similar with some short quick intros what they actually refer to

• What you should do as contributor / how can you fix the CI
• Internals
  • Each check
  • Utilities for each of those, for exampel make style for if only the code quality is failing