Differential-KV (DKV) is a sparse KV-cache inference runtime designed for high-efficiency, memory-bounded long-context Large Language Model (LLM) inference across Apple Silicon (MLX) and CUDA GPUs.
A protocol for computational exhalation: reducing redundant output, compressing stale traces, and preserving reusable intelligence in AI systems.
Deep dive into memory for AI agents: what to store, where to store it, and whether it improves the second run.
金融长文本智能问答系统：检索·判定·记忆·归因四层全链（BM25F 逐选项检索 + 选项级证据隔离 + 分层记忆压缩 + 引用观测归因），自建五域评测集（法规/财报/保险/合同/研报，50 题含 193 页募集说明书），791 例测试。场景口径对齐 AFAC2026 赛题四公开题面，无参赛成绩/官方关联。
Hermes + Yume (夢): sleep-time memory consolidation for Hermes Agent. Replays sessions from state.db and consolidates MEMORY.md / USER.md like a sleeping brain (NREM → REM → Dream Log).
Low-rank eligibility traces for memory-efficient delayed credit assignment.
A memory compression engine built for AI — modular, dual-mode (Compact or Expressive), and part of Nova OS in development. Designed for scalable memory and token efficiency while preserving meaning.
AI agent evaluation system for testing Claude, Gemini & iFlow under memory compression — FastAPI, PostgreSQL, Redis
Compiler-inspired long conversation memory compression: dead memory elimination, fact folding, temporal compression, knapsack pruning
To associate your repository with the memory-compression topic, visit your repo's landing page and select "manage topics."