Update 2:4 Support

[dsikka]

Summary

• Remove compressed support for now
• Validate uncompressed support for tp=1,2,4 for all model combinations
• Add testing cases with nm-testing models:

1. Includes: 2:4 models + FP8 (channel-weight quant, dynamic per token activation quantization)
2. Starting with a base model, applying 2:4, then FP8 (channel-weight quant, dynamic per token activation quantization)
3. Channel weight quant, static per tensor activation quantization
4. Per Tensor weight quant, static per tensor activation quantization
5. Per Tensor weight quant, dynamic per token activation quantization

Models:

• nm-testing/Meta-Llama-3-8B-Instruct-FP8-Dynamic-2of4-testing
• nm-testing/SparseLlama-3.1-8B-gsm8k-pruned.2of4-FP8-Dynamic-testing
• nm-testing/Meta-Llama-3-8B-Instruct-FP8-Static-Per-Tensor-testing
• nm-testing/Meta-Llama-3-8B-Instruct-FP8-Static-testing
• nm-testing/Meta-Llama-3-8B-Instruct-FP8-Dynamic-IA-Per-Tensor-Weight-testing

[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.

🚀