Add hybrid/text search for archival memory

cpacker / MemGPT

Create LLM agents with long-term memory and custom tools 📚🦙

https://memgpt.readme.io

Apache License 2.0

10.77k stars 1.16k forks source link

Add hybrid/text search for archival memory #362

Open sarahwooders opened 8 months ago

sarahwooders commented 8 months ago

Text search might better than low quality embeddings. We can implement another ArchivalMemory (similar to EmbeddingArchivalMemory) that uses text-based/hybrid search.

ohickl commented 8 months ago

I'm interested in achieving as accurate an archival memory as possible, primarily to interact with my scientific literature in a dynamic and intuitive way. Wouldit make sense to be able to fine-tune the quality of the embeddings and adjust the parameters of the generation models? Currently, BAAI/bge-small-en-v1.5 is in use, and I'm not overly familiar with the trade-offs between size and performance. Would upgrading to the larg version notably improve the results?"