HuggingFace每日论文：KV缓存压缩与Agent Harness优化研究
I enjoyed the daily HF papers today
Top 3 papers on HF Daily Paper are all unusually delightful and interesting reads for anyone on the leading edge of local LLMs, agent harness optimization, etc, felt like sharing.
HF Daily Paper 排名前3的论文都异常令人愉悦且有趣，对于任何处于本地LLM、agent harness优化等前沿领域的人来说都是绝佳的阅读材料，忍不住想分享。
DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
DeepSeek-V4.1-Flash：突破KV缓存压缩极限
https://huggingface.co/papers/2609.19969
Cross-layer KV reuse plus FP4 KV caching brings the global KV cache to 890 bytes per token, about a quarter of V4-Flash.
跨层KV复用结合FP4 KV缓存，将全局KV缓存降至每token 890字节，约为V4-Flash的四分之一。
SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness
SoL-Pi：递归扩展自动研究循环以实现高效的Agent Harness
https://huggingface.co/papers/2609.20519
Auto-research loops that improve the agent harness, cutting token traffic by 44.7 to 49.0% at comparable performance.
通过改进agent harness的自动研究循环，在保持相当性能的同时，将token流量减少了44.7%至49.0%。
An Empirical Study of Harness Design for Coding Agents
编码Agent Harness设计的实证研究
https://huggingface.co/papers/2609.20804
Varies planning, action space, and context management across 176 settings to see what each component actually contributes.
在176种设置下变化规划、动作空间和上下文管理，以观察每个组件的实际贡献。
I'm still reading through, feel free to discuss
我还在阅读中，欢迎讨论
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力