-
-
Notifications
You must be signed in to change notification settings - Fork 7.1k
Fix broken tests #14713
New issue
Have a question about this project? No Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “No Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? No Sign in to your account
Fix broken tests #14713
Conversation
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run 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 either: Add 🚀 |
requirements/test.in
Outdated
runai-model-streamer-s3==0.11.0 | ||
|
||
flashinfer-python==0.2.3 # required for flashinfer tests |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
this is installed in the docker. we need to build it. the pypi version is just source code. we cannot install it directly from pypi.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@ProExpertProg can you help take a look at this change to see if it is necessary?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@zou3519 may give a pass as well.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for fixing these tests! Been on my TODO list for a while. Could you also add the test to CI?
tests/compile/test_pass_manager.py
Outdated
if isinstance(callable, InductorPass): | ||
pass_manager.add(callable) | ||
# should succeed for proper InductorPass instances | ||
if works: | ||
pickle.dumps(pass_manager) | ||
else: | ||
with pytest.raises(BypassFxGraphCache): | ||
pickle.dumps(pass_manager) | ||
else: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think this is more complex than it needs to be. The outer if is not necessary, just update the works
boolean if necessary
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the feedback!
Made the changes in this commit: 06b6d03
Done here: b8d0be0 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
seems reasonable to me but I'll defer to @ProExpertProg and @youkaichao
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Upon closer inspection, I realized the test is very outdated. After I added it in #10273, we ended up drastically changing the caching logic. There is no more bypassing the code cache and throwing inside pickle.dumps
. But there is an assertion in pass_manager.add
. I added some suggestions to make it easier to make the change, and thanks again for doing this.
tests/compile/test_pass_manager.py
Outdated
callable_decorated = CallableInductorPass(simple_callable, | ||
InductorPass.hash_source(__file__)) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nit: I would call this callable_uuid
as the uuid is provided manually
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done here: c325dfa
* fix test_pass_manager tests * added imports for test_flashinfer Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
c325dfa
to
684175d
Compare
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com>
@youkaichao I think the CI is failing because it's using torch 2.5 but the precompiled kernels it downloads use 2.6. Why is torch==2.5.1 being used? We could also just force-merge and address later |
@jovsa culd you rebase/merge main for the CI tests? |
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com> Signed-off-by: Louis Ulmer <ulmerlouis@gmail.com>
Signed-off-by: JovanSardinha <jovan.sardinha@gmail.com> Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Fixed some of the broken tests when running
pytest tests/
test_pass_manager
teststest_fash_attn.py