vLLM Major Version Update

vLLM v0.25 stable release adds DeepSeek V4 on AMD, speculative decoding fixes, KV cache offloading, and major performance gains.

v0.25 Stable Release is Here

As usual, here’s a focused summary of the key updates in this release.

New Model Support

This version adds support for quite a few new model architectures:

Special highlight: DeepSeek V4 can now run on AMD GPUs. Great news for anyone who doesn’t want to be locked into NVIDIA.