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
When recalling actions, the agent only consider the similarity to fetch past actions, it will be more correct to integrate the time dimension. We could timestamp each action node when commit in the fact memory, then re-rank the actions in the action retriever. This could be done with unix timestamp. Taking into account the time in the query could also be really useful to recall actions occurred at a particular time (but we needs a proper way to do it).
The text was updated successfully, but these errors were encountered:
When recalling actions, the agent only consider the similarity to fetch past actions, it will be more correct to integrate the time dimension. We could timestamp each action node when commit in the fact memory, then re-rank the actions in the action retriever. This could be done with unix timestamp. Taking into account the time in the query could also be really useful to recall actions occurred at a particular time (but we needs a proper way to do it).
The text was updated successfully, but these errors were encountered: