Chinese contentdair-ai 一周论文精选：HySparse2 与 Harness-Zero 等 7 篇入选
dair-ai 又更新一周论文榜了，这次有 7 篇，稀疏化和智能体自组织团队那几篇可以翻翻，找找自己方向的灵感。
#HySparse2
3 stories · 0 in 7 days · 0 in 30 days
Stories, products and papers tagged "HySparse2", newest first.
Sep 28
Sep 24
Chinese content小米 MiMo 团队发布 HySparse2 注意力架构
小米 MiMo 团队搞了个新注意力架构，1M 上下文时 prefill 计算量和 KV cache 都砍到原来的几分之一，长文检索分数还涨了，做长上下文部署的可以看看论文。
Xiaomi Details HySparse2 Sparse Attention for MiMo-V3
Two-level KV sharing targets the long-context prefill bottleneck that agentic workloads hit, cutting compute without a full attention redesign. Engineers building long-context inference stacks can benchmark the approach against hybrid SWA baselines on similar MoE setups. The MiMo-V3 linkage also signals how Xiaomi positions its model stack against domestic rivals like DeepSeek and Qwen.