·Concavity Team
RAG做不到的事，全文档推理（FCR）做到了！准确率获得4倍提升！
文通过实战案例解析检索增强生成（RAG）在复杂任务中的瓶颈，并介绍新技术——全文档推理如何攻克这些难题。
Read postWe're building Superlinear, an open-source subquadratic attention model that runs locally at 109 tokens/sec with 1M context, and Full Context Reasoning (FCR), a reasoning runtime achieving 4× the accuracy of standard RAG on multi-document tasks.
Early research release. Models and inference engine available.
·Yufeng Huang
Full Context Reasoning (FCR): Moving Beyond RAG for Multi-Document Reasoning
RAG is great for search; it often fails for cross-document reasoning. FCR is a reasoning runtime that constructs a usable full-context environment, then verifies grounding inline as reasoning progresses.
Read post·Yufeng Huang
Introducing Superlinear: Subquadratic Attention for Million-Token Contexts
Attention's quadratic cost is the fundamental bottleneck for long-context inference. We propose Superlinear Multi-Step Attention: a fully trainable attention architecture that achieves O(L^{3/2}) complexity while preserving random context access.
Read post