ci: Add vLLM support to integration testing infrastructure (with qwen) #3545
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
derekhiggins
requested review from
ashwinb,
bbrowning,
ehhuang,
hardikjshah,
leseb,
mattf,
raghotham,
reluctantfuturist,
slekkala1,
terrytangyuan and
yanxi0830
as code owners
meta-cla
bot
added
the
CLA Signed
derekhiggins
force-pushed
the
vllm-ci-qwen
branch
from
e07b4bb to
746e9c9
Compare
slekkala1
reviewed
Oct 8, 2025
derekhiggins
force-pushed
the
vllm-ci-qwen
branch
from
746e9c9 to
5468bab
Compare
derekhiggins
force-pushed
the
vllm-ci-qwen
branch
2 times, most recently
from
88e722c to
1e995fe
Compare
mattf
approved these changes
Oct 9, 2025
| @@ -168,6 +168,11 @@ class Setup(BaseModel): | |||
| roots=base_roots, | |||
| default_setup="ollama", | |||
| ), | |||
| "base-vllm-subset": Suite( | |||
There was a problem hiding this comment.
My intent here was to add this job with only the tests in "tests/integration/inference" and then once we're happy we haven't cause any major disruption we could expand to the entire suit.
There was a problem hiding this comment.
@derekhiggins you feel like this is ready?
There was a problem hiding this comment.
Yes, I believe so,
Although the ollama record job is broken, I've had to rebase with this commit #3898 in order to get new ollama recordings
derekhiggins
force-pushed
the
vllm-ci-qwen
branch
8 times, most recently
from
1c3ea50 to
0f0d986
Compare
Signed-off-by: Derek Higgins <[email protected]>
It preforms better in tool calling and structured tests Signed-off-by: Derek Higgins <[email protected]>
Add vLLM provider support to integration test CI workflows alongside existing Ollama support. Configure provider-specific test execution where vLLM runs only inference specific tests (excluding vision tests) while Ollama continues to run the full test suite. This enables comprehensive CI testing of both inference providers but keeps the vLLM footprint small, this can be expanded later if it proves to not be too disruptive. Also updated test skips that were marked with "inline::vllm", this should be "remote::vllm". This causes some failing log probs tests to be skipped and should be revisted. Signed-off-by: Derek Higgins <[email protected]>
Signed-off-by: Derek Higgins <[email protected]>
derekhiggins
force-pushed
the
vllm-ci-qwen
branch
from
0f0d986 to
6e65e72
Compare
The vector_provider_wrapper was only limiting providers to faiss/sqlite-vec for replay mode, but CI tests also run in record mode with the same limited set of providers. This caused test failures when trying to test against milvus, chromadb, pgvector, weaviate, and qdrant which aren't configured in the record job.
derekhiggins
force-pushed
the
vllm-ci-qwen
branch
from
83a73b1 to
b4e928c
Compare
nathan-weinberg
approved these changes
Oct 28, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
o Introduces vLLM provider support to the record/replay testing framework
o Enabling both recording and replay of vLLM API interactions alongside existing Ollama support.
The changes enable testing of vLLM functionality. vLLM tests focus on
inference capabilities, while Ollama continues to exercise the full API surface
including vision features.
--
This is an alternative to #3128 , using qwen3 instead of llama 3.2 1B appears to be more capable at structure output and tool calls.