vLLM v0.24.0 adds MiniMax-M3 support and performance improvements.

Core summary

One sentence to understand this update

vLLM v0.24.0 has been released, featuring 571 commits from 256 contributors and notable support for the new MiniMax-M3 model, alongside BF16/FP8 indexer enhancements.

Impact & opportunity

What this could mean

Developers using vLLM can now leverage MiniMax-M3 and benefit from performance improvements, expanding their options for deploying and serving large language models.