AIMarkTechPostabout 1 hour ago
Stanford Releases Frozen LLM Memory System
1 min read
Stanford researchers released Memorilla, an open-source memory system for LLMs that reads whole document collections and hands frozen models just 16 memory tokens. On FactKG, a frozen Qwen3-8B scored 84.33% with Memorilla against 80.43% with every document in context, achieving this with 108x fewer document tokens. The code, training recipes and checkpoints are public under Apache 2.0.
Level
Hype check
Tap to vote and see what everyone thinks.
