[Bugfix][Kernel] Fix per-token/per-channel quantization for Hopper scaled mm

[tlrmchlsmth]

This fixes the selection between grouped vs per-tensor or per-token quantization. In some cases, the linear layers will pass in transposed scales (e.g. 1 x num_tokens instead of num_tokens x 1). This simplifies and makes the dispatching to the per-token/per-channel/per-tensor case more flexible.

Fixes the error:

E       RuntimeError: Unsupported scale group shapes for CUTLASS 3.x GEMM.
E        a_scale_group_shape must be [1, 128], got: [1, 2048]
E       b_scale_group_shape must be [128, 128], got: [1, 2560]

when running for example: vllm serve nm-testing/TinyLlama-1.1B-Chat-v1.0-FP8_DYNAMIC-e2e

[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.

🚀

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @tlrmchlsmth.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[tlrmchlsmth]

This is the main change

[LucasWilkinson]

LGTM, thanks for fixing this!

[mgoin]

LGTM