vLLM adds Triton ReplaySSM decoding and Model Runner V2
November 3, 2026
vLLM has introduced Triton ReplaySSM decoding on the GPU and support for Model Runner V2. These updates aim to optimize decoding performance and architectural modularity.
HOW THIS AFFECTS YOU
●
builderYou can achieve faster decoding performance by utilizing the new Triton ReplaySSM implementation.