devwithjevUses Jev to Rank AtMem Memories
AtMem tested Jev 1.13.0 on the official LoCoMo data benchmark, where it improved the rate of the correct memory appearing first from 33.99% to 54.23%.
View on X costcosted about $0.0583
We did a test using Jev 1.13.0 on official LoCoMo data benchmark.
Jev makes AtMem better at putting the right memory first, but it does not yet help AtMem find memories it missed entirely.
The significant increase is on how often the correct memory was the very first result. Improved from 33.99% to 54.23%
AtMem used about 1,593,522 token and costed about $0.0583
Jev remains as option in the next releases since AtMem Position is local first.
We continue to explore more use cases for Jev
Jev improved memory ordering but did not help AtMem find memories it had missed entirely. AtMem plans to keep Jev as an option in future releases because AtMem Position is local first.
Also filed under Tools & apps
- Swift Client for TypeSafe Jev Judgments
swift-jev is a Swift client for TypeSafe Jev that uses typed judgments instead of generated text.
- Go SDK and CLI for Jev Decisions
go-jev provides a Go SDK and CLI for TypeSafe Jev typed yes/no, Choice, and Score decisions.
- Open-Model Semantic-If Engine
SemIf-OpenJev is an open-model semantic-if engine for local Jev-style decisions.
- Open-Weight Jev Decision Model
JevK5 is an open-weight Jev alternative that returns typed decisions with probabilities in one forward pass.