EverMind

EverMind Releases Memory-Sparse Attention Code for Long-Context AI Models

EverMind has open-sourced its "Memory-Sparse Attention" (MSA) system, allowing AI models to process up to 100 million tokens with minimal performance degradation. This innovative system integrates memory directly into the model's architecture, outperforming larger RAG systems in benchmarks. The release has garnered significant industry interest.