[Bugfix][Model] OLMo 2: split qkv correctly for GQA and MQA

[2015aroras]

Issue: OLMo 2 does not correctly do attention when the number of heads is not the same as the number of kv heads (i.e. GQA or MQA is used instead of MHA). Specifically, it splits qkv into equal chunks rather than chunks based on q, k, v size.

Fix: Split qkv based on q, k, v size.

Testing: We ran this internally on a model that uses GQA.

FIX #13686

[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.

🚀