Reranker?

[sysadmvoron-sys]

Sorry guys, but honestly i dont understand where Reranker is? I can't even build a basic RAG because i cannot attach my local reranker just like I did with emb and vlm itself.

Do i miss something? How it's possible that service that have "everything" to build a good RAG or whatever missing most cruvial middle tool called "reranker"? Is this an vehicle without motor or do I really miss something?

[dasu07988]

Hi! I understand your confusion. I checked the current Flowise version, and reranking functionality is available through Cohere Rerank Retriever and Voyage AI Rerank Retriever.

However, I think the main issue you're referring to is the lack of a generic/local reranker integration. If you're trying to use a self-hosted reranker (for example, a Hugging Face CrossEncoder, BGE Reranker, or another local service), Flowise doesn't currently provide a built-in node where you can simply plug in any local reranker the same way you can configure embedding models or LLMs.

As a workaround, you could:

Integrate your local reranker through a custom node.
Expose the reranker as an API and call it using an HTTP Request node.
Perform the reranking in your application before passing the final documents to the LLM.

If your goal is to build a fully local RAG pipeline, native support for custom/local rerankers would definitely make Flowise more flexible.

Just to clarify, are you looking for support for a specific reranker (such as BGE Reranker, Jina AI Reranker, NVIDIA NIM Reranker, or another local model)? That would help determine the best integration approach.

[dasu07988]

Hi! I understand your confusion. I checked the current Flowise version, and reranking functionality is available through Cohere Rerank Retriever and Voyage AI Rerank Retriever.

However, I think the main issue you're referring to is the lack of a generic/local reranker integration. If you're trying to use a self-hosted reranker (for example, a Hugging Face CrossEncoder, BGE Reranker, or another local service), Flowise doesn't currently provide a built-in node where you can simply plug in any local reranker the same way you can configure embedding models or LLMs.

As a workaround, you could:

Integrate your local reranker through a custom node.
Expose the reranker as an API and call it using an HTTP Request node.
Perform the reranking in your application before passing the final documents to the LLM.

If your goal is to build a fully local RAG pipeline, native support for custom/local rerankers would definitely make Flowise more flexible.

Just to clarify, are you looking for support for a specific reranker (such as BGE Reranker, Jina AI Reranker, NVIDIA NIM Reranker, or another local model)? That would help determine the best integration approach.