Milvus Model
Milvus's model library with a Jev reranker that sends candidate documents as yes/no relevance questions and returns them sorted.
Overview
milvus-model is the library that integrates common embedding and reranker models with Milvus, the open-source vector database, and is included as an optional dependency of pymilvus. It supports providers such as OpenAI, Voyage AI, Cohere, Jina AI and TypeSafe (Jev), plus open-source models through SentenceTransformers or Hugging Face's Text Embeddings Inference.
Its Jev integration is a Python reranker adapter: it sends each candidate document as a noul relevance question against the query, batched in one request, then sorts the documents by returned score while preserving their original indices. The README excerpt covers installation rather than the reranker's details, which come from the listings. It suits Milvus users who want to add a typed relevance judgment as a second-stage reranker without hosting a cross-encoder.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this candidate document relevant to the query? | — |
How it works
- Retrieve candidate documents from Milvus
- Pass query and candidates to the Jev reranker
- Jev answers a relevance question per document in one request
- Documents are sorted by score with original indices kept
Strengths
- Single request for all candidates
- Fits into existing pymilvus workflows
- No local reranker model to host
Caveats
- Requires a TypeSafe key
- Reranker details not in the main README
- Documents sent to the hosted API
Getting started
pip install pymilvus[model] (or pymilvus.model) and configure the TypeSafe reranker.At a glance
- Type
- Library
- Status
- production
- Language
- Python
- Runtime
- library (pymilvus extra)
- Has code
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
Search infrastructure: batches candidate-document Noul questions through Jev and returns score-sorted results with original indices through a Python reranker adapter.
yibie/awesome-jevPython reranker adapter that sends candidate documents as Jev Noul questions in one request, then sorts the returned scores and preserves original document indices
AnotiaWang/awesome-decision-modelsA library integrating embedding and reranker models from OpenAI, SentenceTransformers etc for semantic search in vector database.
logicrw/awesome-jev-projectsPython reranker adapter that sends candidate documents as Jev Noul questions in one request, then sorts the returned scores and preserves original document indices.
cobanov/awesome-jevPython reranker adapter that sends candidate documents as Jev Noul questions in one request, then sorts the returned scores and preserves original document indices.
hellogumbo/awesome-jev
Frequently asked questions
How does the Jev reranker work?
It asks one yes/no relevance question per document in a single request and sorts by probability.
How do I install it?
pip install pymilvus[model].
Which other providers are supported?
OpenAI, Voyage AI, Cohere, Jina AI and open-source models, among others.
Related projects
Drop-in rerank API that speaks Cohere, Jina and Voyage request shapes and scores each document with a Jev relevance question.
ServicebetaJavaScriptRails gem for intent search: Jev answers declared questions when records are saved, and searches become weighted SQL over those answers.
ProjectbetaRubyUnofficial LlamaIndex reranker and router where Jev scores retrieved passages and selects query engines, with BEIR results.
LibrarybetaPythonRubyLLM provider gem for TypeSafe Jev: block-style typed evaluations with noul, choice and score, plus RubyLLM.rerank support.
ProjectbetaRuby248 small installable TypeScript recipes that each make one typed Jev decision: route, rerank, gate, grade, compare or label.
LibraryproductionTypeScriptJava SDK for TypeSafe's System One API plus Spring AI integrations using Jev as judge, guardrail, RAG filter/reranker and tool index.
LibrarybetaJava