[Bugfix] Fix Mamba model initialization and MLP Speculator weights loading

[Isotr0py]

FIX #10327 (comment)

Sorry for breaking the test :(

[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.

🚀

[Isotr0py]

@DarkLight1337 Seems that the failing Mamba loading is not a bug from weights loading tracker: https://buildkite.com/vllm/ci-aws/builds/11414#019342cb-092d-405f-8d84-89c9f345edd8/182-3271. So I disabled Mamba weights loading tracking temporarily.

I indeed can't find RMS norm in the model checkpoint: https://huggingface.co/tiiuae/falcon-mamba-tiny-dev/tree/main?show_file_info=model.safetensors, perhaps the tested mamba model was not initialized correctly.

[tlrmchlsmth]

Thanks for the fix!