[CB] Update CB docs + Refactoring scheduling step-by-step inference tests

[sducouedic]

Two main changes in this PR:

• Scheduling tests refactoring: convert previously unit-test test_scheduler_cb_steps_tkv to a helper function, and the previously parameters functions of test_scheduler_cb_steps_tkv to unit tests
• Build doc from tests docstring (continuous batching only), check these pages:
  • Doc e2e tests
  • Doc stepping tests
  • Other CB tests

Closes #318
Closes #319

[github-actions]

👋 Hi! Thank you for contributing to vLLM support on Spyre.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, first install the linting requirements, then run format.sh and commit the changes. This can be done with uv directly:

uv sync --frozen --group lint --active --inexact

Or this can be done with pip:

uv pip compile --group lint > requirements-lint.txt
pip install -r requirements-lint.txt
bash format.sh

Now you are good to go 🚀

[sducouedic]

I just noticed that now the cb scheduling tests are skipped, trying to figure out why is this 🤔

[sducouedic]

I just noticed that now the cb scheduling tests are skipped, trying to figure out why is this 🤔

Oh I found out, I didn't know unit tests had to be prefixed with test_. I just learned something 😂

[sducouedic]

thanks for being so reactive in your review for this big PR @prashantgupta24 , I'll finish addressing your comments tomorrow

I see that test_openai_serving_cb is failing, I am not sure if it is related to those changes or if it is something related with multiprocessing, it was already failing in a previous commit and then passing again...

[yannicks1]

great restructuring Sophie! left a couple of minor comments.

[yannicks1]

Can we find a better title for this section? It would be nice to coincide with the chapter on the left of the page. Currently we have:

End-to-End -> title: End output correctness tests -> file name: e2e_testing.md
Step-by-Step -> title: Scheduler inference steps tests -> file name: scheduling_inference_steps.md
Other Tests -> title: Other tests -> file name: other_tests.md

what about:

Output Tests -> title: Output Tests -> file name: output_tests.md
Scheduler Tests -> title: Scheduler Tests -> file name: scheduler_tests.md
Other Tests -> title: Other tests -> file name: other_tests.md

e.g. making things more uniform?

[sducouedic]

I am not sure about scheduler_tests only because there is some other test test_cb_max_tokens in test_spyre_cb.py that is also testing the scheduler (but just the constraint), without really belonging to this test suite imo. I want to emphasis that this is somehow "step-by-step correct execution". And it won't be only about the scheduler correctness, on the compiler side it is also required that the compiled graph cab adapt to different scenarios of prompts_length/max_tokens/steps joining.

But I agree the current naming is a bit fuzzy, I changed it a lot

[sducouedic]

But output_test is great naming, I change that

[sducouedic]

For the scheduler tests, I renamed everything to scheduler_steps tests 👍

[prashantgupta24]

Last question of whether we want to integrate the test_cb_max_tokens with test_spyre_basic, otherwise I can approve!

[prashantgupta24]

Nice!