You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As outlined in the readme, memary currently extracts the top N entities (set to 20) from the Entity Knowledge Store. We pass these entities into the new context window for inference.
memary can future benefit from more advanced memory compression techniques such as passing only the entities that are present in the agent's response to the context window. This will help the system understand the users' knowledge of concepts for that specific execution which will lead to more relevant responses.
We look forward to related community contributions.
The text was updated successfully, but these errors were encountered:
If the response from the ReAct agent is “The Mona Lisa can be found in the Louvre in Paris” then only the counts and recency of these entities (Mona Lisa, Louvre, Paris) are passed into the context window to infer user knowledge around the topic. Before we were passing the top 20 entities regardless, not considering which entities were in the response.
As outlined in the readme, memary currently extracts the top N entities (set to 20) from the Entity Knowledge Store. We pass these entities into the new context window for inference.
memary can future benefit from more advanced memory compression techniques such as passing only the entities that are present in the agent's response to the context window. This will help the system understand the users' knowledge of concepts for that specific execution which will lead to more relevant responses.
We look forward to related community contributions.
The text was updated successfully, but these errors were encountered: