[CI][Spec Decode] fix: broken test for EAGLE model

[llsj14]

#11672 (comment)
here was an issue in the above PR that caused broken tests and a CI test omission.
This PR fixes the broken tests and updates the test pipeline.

test

As follows, broken unit tests were passed in my local environments.

pytest ./tests/spec_decode/e2e/test_eagle_correctness.py
========================================= test session starts ==========================================
platform linux -- Python 3.10.12, pytest-7.4.2, pluggy-1.3.0
rootdir: /mnt/lvm/sungjae/pr/vllm
configfile: pyproject.toml
plugins: asyncio-0.23.7, forked-1.6.0, anyio-3.7.1, hypothesis-5.35.1, flakefinder-1.1.0, rerunfailures-12.0, shard-0.1.2, xdist-3.3.1, xdoctest-1.0.2
asyncio: mode=strict
collected 15 items                                                                                     
Running 15 items in this shard

tests/spec_decode/e2e/test_eagle_correctness.py ...............                                  [100%]
=========================================== warnings summary ===========================================
…
-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
============================= 15 passed, 17 warnings in 482.77s (0:08:02) ==============================

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[llsj14]

With the standard EAGLE model checkpoint, this part is not necessary. However, the EAGLE model ("abhigoyal/vllm-eagle-llama-68m-random") used in spec_decode/e2e/test_eagle_correctness.py contains an input layernorm in its checkpoint. If I simply remove the weight in this DummyInputLayerNorm, it could cause errors when loading the model's weights.

[sroy745]

nit: Can you please add a comment on the need to have these weights for some of the unit tests that we have?

[llsj14]

@DarkLight1337
I just fixed the issues mentioned in this comment. Please take a look.

[DarkLight1337]

Thanks for the quick fix! To avoid starting up a new CI instance just for this test, can you combine this with the speculative decoding CI?

[llsj14]

Thanks for the quick fix! To avoid starting up a new CI instance just for this test, can you combine this with the speculative decoding CI?

I was confused whether it was good or not to combine. Thank you for your feedbacks. I just updated it as you recommended.

[sroy745]

Thanks for the fix.