WIP (based on #485, #489): Rewrite declarative memory #492
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
DeclarativeMemory was sensitive to insertion order, making batch ingestion difficult to implement. Now episodes are only connected by timestamp during search.
Implement batch ingestion: LoCoMo ingestion time reduced from 5-10 minutes to 30-40 seconds (huge improvement!) from cold start.
DeclarativeMemory configuration is too complicated for most people. Make it into a black box to make it easier to use and get some extra performance.
Returning chunks instead of episodes allows future implementation of chunk size limits. Previously there was clustering, which only made things bigger.
Security logic was easy to compromise in the case of a poor DerivativeDeriver, DerivativeMutator, or RelatedEpisodePostulator implementation. Blackboxing makes security easier to reason about.
Data type changes:
Ingestion/search changes:
Type of change
How Has This Been Tested?
Manually tested on LoCoMo dataset (updated scripts not committed yet), agentless.
TODO: Unit/integration tests. Update evaluation scripts.
Checklist
Maintainer Checklist