[GH-ISSUE #3942] [FEAT]: Allow Free Selection of Reranking Models to Improve Support for Non-English Languages #2510

Closed
opened 2026-02-22 18:30:00 -05:00 by yindo · 4 comments
Owner

Originally created by @Sarmingsteiner on GitHub (Jun 3, 2025).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3942

What would you like to see?

Description:
I would like to request that AnythingLLM enables free selection and integration of different reranking models, especially to enhance the quality of search and ranking results for languages other than English.

Background:
Many existing reranking models are primarily optimized for English and tend to perform worse on queries and documents in other languages (e.g., German, Spanish, French). Allowing users to choose or integrate language-specific reranking models would significantly improve relevance and precision in multilingual use cases.

Use Cases:

Organizations and users working primarily in non-English environments

Applications handling multilingual documents and queries

Flexibility to swap or add reranking models tailored for specific languages or domains

Suggested Implementation:

Make reranking models modular and configurable to allow easy replacement or addition

Support automatic selection or manual configuration of reranking models based on language or context

Provide documentation and examples for integrating custom or external reranking models

Benefits:

Better search and ranking accuracy for non-English languages

Increased flexibility and adaptability for international and multilingual deployments

Improved user experience and broader applicability of AnythingLLM globally

Thank you for considering this enhancement!

Originally created by @Sarmingsteiner on GitHub (Jun 3, 2025). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/3942 ### What would you like to see? **Description:** I would like to request that AnythingLLM enables free selection and integration of different reranking models, especially to enhance the quality of search and ranking results for languages other than English. **Background:** Many existing reranking models are primarily optimized for English and tend to perform worse on queries and documents in other languages (e.g., German, Spanish, French). Allowing users to choose or integrate language-specific reranking models would significantly improve relevance and precision in multilingual use cases. **Use Cases:** Organizations and users working primarily in non-English environments Applications handling multilingual documents and queries Flexibility to swap or add reranking models tailored for specific languages or domains **Suggested Implementation:** Make reranking models modular and configurable to allow easy replacement or addition Support automatic selection or manual configuration of reranking models based on language or context Provide documentation and examples for integrating custom or external reranking models **Benefits:** Better search and ranking accuracy for non-English languages Increased flexibility and adaptability for international and multilingual deployments Improved user experience and broader applicability of AnythingLLM globally Thank you for considering this enhancement!
yindo added the enhancementfeature request labels 2026-02-22 18:30:00 -05:00
yindo closed this issue 2026-02-22 18:30:00 -05:00
Author
Owner

@learning4life commented on GitHub (Jun 5, 2025):

We like to see this added 👍

@learning4life commented on GitHub (Jun 5, 2025): We like to see this added 👍
Author
Owner

@Alminc91 commented on GitHub (Jul 5, 2025):

Will it be possible to host the reranker model on my GPUs and let AnythingLLM reference it via OpenAI compatible? That would be perfect. In our system, the reranker doubles the retrieval time. It could be because the CPU is not the best and most powerful.

@Alminc91 commented on GitHub (Jul 5, 2025): Will it be possible to host the reranker model on my GPUs and let AnythingLLM reference it via OpenAI compatible? That would be perfect. In our system, the reranker doubles the retrieval time. It could be because the CPU is not the best and most powerful.
Author
Owner

@timothycarambat commented on GitHub (Jul 7, 2025):

@Alminc91 - that is what we would provide. it would function just like LLMs or Embedders, where you specify the provider and then the model tag so we can send a request. Generic OpenAI compatible systems will be a provider.

This gives you the ability to run services on device that are more hardware optimized (like on GPU) because yes CPU can be super limiting and annoying when dealing with larger reranking sets

@timothycarambat commented on GitHub (Jul 7, 2025): @Alminc91 - that is what we would provide. it would function just like LLMs or Embedders, where you specify the provider and then the model tag so we can send a request. Generic OpenAI compatible systems will be a provider. This gives you the ability to run services on device that are more hardware optimized (like on GPU) because yes CPU can be super limiting and annoying when dealing with larger reranking sets
Author
Owner

@timothycarambat commented on GitHub (Sep 22, 2025):

Moving into #4012

@timothycarambat commented on GitHub (Sep 22, 2025): Moving into #4012
yindo changed title from [FEAT]: Allow Free Selection of Reranking Models to Improve Support for Non-English Languages to [GH-ISSUE #3942] [FEAT]: Allow Free Selection of Reranking Models to Improve Support for Non-English Languages 2026-06-05 14:46:59 -04:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#2510