Doradus ResearchOperator-perspective notes on building AI infrastructure.https://doradusresearch.ai/Frontier MoE sleep/wake at TP=4 on consumer Blackwell — 1s wake, 3s swaphttps://doradusresearch.ai/blog/vllm-blackwell-sleep-mode/https://doradusresearch.ai/blog/vllm-blackwell-sleep-mode/DeepSeek-V4-Flash + MiMo-V2.5-Flash rotation on 4× RTX PRO 6000. Cross-peer swap 50s down to 3s. The sparse-MLA workspace, the SWA release bug, the cumem race, and the one config knob that made it work.Sat, 16 May 2026 07:00:00 GMT