MEMENTO teaches 8B model to manage long-context notes
Researchers trained Qwen3-8B to compact long books into notes, lifting book QA accuracy to 67.8% in a 20K-token window. Here's how it works in practice.
Try in Readermementolong contextcontext compactionqwen3-8b