The Platform Development team has been busy working away for some time on a new, faster engine that we are now excited to put live as our primary inference engine. This is being deployed in our v4.1 release.
This has been available to users under the “beta” moniker since last year and has been extensively tested over the past 6 months. Every query made during this time went to both old and new engines, with all responses being compared for accuracy and mismatches analysed by the development team. This has ensured the accuracy of the inference engine is unaffected by this update.
This update means you can now:
Get faster query results: faster inferencing means less waiting time for the next question or your result. Our tests have shown the new engine performs around 90% faster.
Provide more data, more efficiently: It can support more facts, proving an ability to process 1000 injected facts 99% faster than the previous engine. This means more information can be provided to the engine when running a query to support fully-autonomous use cases.
Deepen your search space: The new inference engine has a query depth of 2,000,000 sub-queries. This greatly reduces the risk of combinatorial explosion making your queries fail. Considering the old inference engine was limited to 10,000 sub-queries, this gives you much more room for complexity.
Following the go-live, this new engine will be used for all queries, unless explicitly stated otherwise.
Important information following this engine upgrade:
- Some old knowledge maps that have not been updated since August 2020 will not run. We have identified and contacted as many users as possible that need to resave their maps. However this can be done by opening your knowledge map in the Studio and saving it. This will save your knowledge map in the correct format required by the new engine.
- Linked knowledge maps and compounds are engine features not currently supported by this new engine. These can continue to be used in the old (v1.0) engine, which will remain available for a period of time.
- Both Engines can still be selected within the Studio.
- In Test Settings, the selection will now default to v2.0, our new engine. However v1.0, our legacy engine, can still be accessed.
- Within an Agent configuration, both v1.0 and v2.0 engines are selectable
- When using our API, it will default to v2.0, however a header can be used to explicitly request v1.0 if it is required. Details of this can be found here: https://rainbird-ai.gitlab.io/rainbird-docs/#22-experimental-and-alternate-engines
We hope you enjoy this turbo-charged engine that can work faster and think deeper. If you have any questions, comments or issues, please feel free to reach out to us here.
The Platform Team