vLLM Implements FlashMLA and NVFP4 Compressed KV Cache

vLLM has integrated FlashMLA mega attention and support for NVFP4 compressed KV caches. These optimizations significantly reduce memory footprint and increase throughput during long-context inference.