Victorwz/LongMem
A research project implementing a NeurIPS 2023 paper that augments language models with long-term memory using a side network and memory bank retrieval.
Not currently ranked — collecting fresh signals.
star history
This project implements a language model architecture capable of utilizing long-term memory. It augments a frozen pre-trained LLM backbone with a trainable side network and joint attention mechanism to retrieve and integrate relevant past context from a memory bank. The system uses Faiss-GPU for efficient vector similarity search over encoded memory chunks, enabling the model to attend to information from extended context windows.
Frequently asked
- What is Victorwz/LongMem?
- A research project implementing a NeurIPS 2023 paper that augments language models with long-term memory using a side network and memory bank retrieval.
- Is LongMem open source?
- Yes — Victorwz/LongMem is open source, released under the Apache-2.0 license.
- What language is LongMem written in?
- Victorwz/LongMem is primarily written in Python.
- How popular is LongMem?
- Victorwz/LongMem has 827 stars on GitHub.
- Where can I find LongMem?
- Victorwz/LongMem is on GitHub at https://github.com/Victorwz/LongMem.