DeepSeek V4

vLLM v0.25.0 introduces Model Runner V2 as the default for dense models, significantly improving performance and adding support for new features like EVS and realtime embeddings. The release also deprecates PagedAttention and enhances the Transformers backend to match native vLLM speed, alongside numerous model additions and performance optimizations across various hardware platforms.

This release focuses heavily on DeepSeek V4 maturity with new kernel support and packaging, significant advancements in Model Runner V2, and the introduction of an experimental Rust frontend. Performance saw notable gains from batch-invariant inference with Cutlass FP8 and the rollout of multi-tier KV cache offloading.

This release introduces significant performance and stability improvements, notably integrating KV offloading with the Hybrid Memory Allocator and enabling speculative decoding with thinking budgets. It also formally deprecates support for older versions of the Transformers library.

vLLM v0.20.1 is a patch release focused on stabilizing and improving performance for DeepSeek V4, including various kernel optimizations and critical bug fixes across CUDA and ROCm platforms.