-
Type:
Epic
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: ABX, ai-ml-testing, LangChain, LangGraph
-
None
-
Native Search Result Reranking in AI Framework Integrations
-
Python Drivers
-
Not Needed
-
-
In Progress
-
None
-
0
-
0
-
0
-
100
-
None
-
None
-
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Per DRIVERS-3301:
The cloud-qa search cluster (drivers-testing project) now supports $rerank — the MMS version has been updated and the project setting is enabled. Teams can run rerank queries against it. That cluster is currently shared by java team and csharp team for atlas search tests. Cloud-dev should also be working if anyone wants to use a cluster running 8.3+ there instead.
Note on models: rerank-2.5-lite is the only model currently backed by real GPUs. Other models (rerank-2.5, rerank-2, rerank-2-lite) return constant scores.
The PD for the Public Preview has requested
While it is already supported to use Voyage Rerankers with AI Frameworks directly through existing Voyage AI integration (e.g. LangChain), we must expose the benefits of $rerank as native reranking functionality directly into AI Frameworks through their existing MongoDB search integrations:
The experience should mimic LangChain’s semantic reranker option for Azure AI Search, ie.
Specify AzureAISearchQueryTypeConst as “SemanticHybrid”.