vLLM adds top_k and top_p sampling to DiffusionGemma
July 26, 2026
vLLM now supports top_k and top_p sampling parameters for DiffusionGemma models. This enables more granular control over stochasticity during the image generation inference process.
HOW THIS AFFECTS YOU
●
builderYou can now implement more precise sampling controls for DiffusionGemma in production.
●
designerThis provides better control over the randomness and variety of generated visual outputs.