[CI Bugfix] Fix CI OOM for test_shared_storage_connector_hashes

[mgoin]

Purpose

This test has been consistently failing since its addition in #21611
https://buildkite.com/vllm/ci/builds/25475/steps/canvas?sid=01985c5a-a224-435b-ab54-41cb1837705a#01985c5a-a380-4372-bf58-7de2780ad329/7-13621

History https://buildkite.com/organizations/vllm/analytics/suites/ci-1/tests/7b01abb7-8064-8f64-9362-7eceb5b9280e?period=7days

Test Plan

Green CI

Test Result

[github-actions]

👋 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 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 either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Code Review

This pull request aims to fix a recurring out-of-memory (OOM) error in the test_shared_storage_connector_hashes CI test. The changes introduce gpu_memory_utilization=0.4 and enforce_eager=True to the EngineArgs for this specific test. This is a sound approach to reduce memory consumption in a resource-constrained CI environment, as it limits the KV cache size and disables CUDA graphs which can be memory-intensive. The changes are localized to the test and appear correct.

[DarkLight1337]

Thanks for fixing

[tlrmchlsmth]

Looks like this model also breaks the CI unfortunately:

[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683] ValueError: Failed to find a kernel that can implement the WNA16 linear layer. Reasons:
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683] MacheteLinearKernel requires capability 90, current compute  capability is 89
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683]  AllSparkLinearKernel cannot implement due to: For Ampere GPU, AllSpark does not support group_size = 128. Only group_size = -1 are supported.
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683]  MarlinLinearKernel cannot implement due to: Weight output_size_per_partition = 6840 is not divisible by  min_thread_n = 64. Consider reducing tensor_parallel_size or running with --quantization gptq.
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683]  Dynamic4bitLinearKernel cannot implement due to: Only CPU is supported
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683]  BitBLASLinearKernel cannot implement due to: bitblas is not installed. Please install bitblas by running `pip install bitblas>=0.1.0`
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683]  ConchLinearKernel cannot implement due to: conch-triton-kernels is not installed, please install it via `pip install conch-triton-kernels` and try again!
[2025-08-03T23:44:26Z] (EngineCore_0 pid=14405) ERROR 08-03 16:44:26 [core.py:683]  ExllamaLinearKernel cannot implement due to: Exllama only supports float16 activations

[tlrmchlsmth]

Did we change the device running this test?