vLLM v0.24.0 update includes MoE refactor and Qwen3 NVFP4 configurations

Core summary

One sentence to understand this update

vLLM has released version 0.24.0, featuring a refactor for Mixture-of-Experts (MoE) models and configurations for Qwen3 NVFP4.

Impact & opportunity

What this could mean

This release likely improves the efficiency and compatibility for deploying MoE models and Qwen3 on NVIDIA GPUs with FP4 precision, benefiting model serving.