[Quant] Fix use_mla TypeError and support loading pure-sparsity Compressed Tensors configs

[kylesayrs]

Purpose

• Fix type error in use_mla which was causing models with quantization and without config_groups to fail
• Support loading models which do not have config_groups in the compressed tensors config. While these models aren't accelerated, they can still be generated from valid recipes in LLM Compressor and are useful for testing.

Changes

• Fix type error in use_mla
• Add warning and return no scheme if the model is attempted to load without quantization

Testing

• Tested nm-testing/Qwen2-VL-2B-Instruct-Sparse-0.6
  • Confirmed that MLA is indeed disabled in the fallback case

WARNING 02-03 16:37:36 config.py:1008] compressed-tensors MLA support requires fp8 activations and weights in group 'group_0', but got activations type 'None' and weights type 'int'.

• Tested neuralmagic/Sparse-Llama-3.1-8B-ultrachat_200k-2of4-quantized.w4a16

[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.

🚀