Publisher · verified by editors

vLLM Blog

AI news source. Articles are auto-selected and adapted by Hamidun News editors.

2 articles in Hamidun·Latest: October 12· Active·vllm.ai ↗
Subscribe

Latest publications

vLLM Teaches Decode Nodes to Send KV Cache Back, Cutting TTFT by Up to 3.3x
LLMvLLM Blog

vLLM Teaches Decode Nodes to Send KV Cache Back, Cutting TTFT by Up to 3.3x

vLLM now supports bidirectional KV cache transfer: prefill nodes pull precomputed context from decode nodes, cutting time to first token on long conversations by up to 3.3x.

Oct 12, 2026·3 min
vLLM Now Runs on Vera Rubin NVL72: 7.8x Higher Per-GPU Throughput
LLMvLLM Blog

vLLM Now Runs on Vera Rubin NVL72: 7.8x Higher Per-GPU Throughput

vLLM now supports NVIDIA Vera Rubin NVL72 with day-0 model support and Rubin-tuned FlashInfer kernels. On AgentX, per-GPU throughput is 7.8x that of GB200.

Oct 10, 2026·3 min