06:06
2026-07-07
mimo.xiaomi.com
large-language-models
Inference Optimization for MiMo v2.5: Pushing Hybrid SWA Efficiency to the Limit
The MiMo-V2.5 model family introduces Hybrid Sliding Window Attention (Hybrid SWA) to reduce KVCache storage to roughly 1/7 of Full Attention, sparse MoE activation to cut per-token compute, and multi…