Memorilla releases code for compact semantic memory in LLMs
The Memorilla developers released Apache 2.0-licensed code, checkpoints, and evaluation recipes for condensing document collections into a fixed set of query-dependent memory tokens for frozen LLMs. They report that 16 memory tokens achieved 84.33% FactKG accuracy versus 80.43% for full-document prompting while using 108 times fewer tokens, and raised average TextWorld agent success from 18.6% to 24.2%. The repository also includes reinforcement-learning workflows based on TextWorld and GRPO.
Memorilla provides an implementable approach to giving frozen LLMs compact access to large document collections, with reported token-efficiency and benchmark gains.
🦍 Memorilla: Latent Semantic Memory for LLMs

Memorilla: Latent Semantic Memory for LLMs
This repository provides code for training and evaluating Memorilla, a memory system that gathers information from an entire document collection into a fixed number of query-dependent memory tokens for a frozen LLM. The code is based on our paper Memorilla: Latent Semantic Memory for LLMs.
#Memorilla#Qwen3-8B#Qwen3-Embedding-4B#Mem0#MemoRAG#FactKG#TextWorld#SkyRL-Memorilla