NVIDIA Component Model Reset to Default Value Issue

[andhwynm]

Bug Description

An error occurred when using the NVIDIA Rerank component in LangFLow v1.9.1.

I have investigated the root cause. The selected paid model is not supported by the current API key.

Both the NVIDIA Rerank and Embedding components have an issue where the selected model resets to the default value.

Reproduction

1. The open-source model I selected【nv-rerank-qa-mistral-4b:1】 worked successfully during initial execution.
   However, after closing and reopening the Flow, the model reverted to the default value【nvidia/llama-3.2-nv-rerankqa-1b-v1】.

2, In Code view, the model still shows as the model 【nv-rerank-qa-mistral-4b:1】.

Expected behavior

How to keep the selected model unchanged in the NVIDIA Rerank component

Who can help?

No response

Operating System

Window11

Langflow Version

Langflow v1.9.1 ( Install and run the Langflow OSS Python package )

Python Version

3.12

Screenshot

No response

Flow File

PGVector RAG.json

[myshkin451]

Hi, I opened a focused PR for this: #13428
It preserves the selected NVIDIA rerank/embedding model when refreshed model options still include it, with fallback behavior covered by unit tests.

[andhwynm]

Thanks a lot for your reply.
Look forward to the fix for this issue being included in the next release.

We are currently using the following alternative solutions:
Replace NVIDIA Rerank with Cohere Rerank
Replace NVIDIA Embedding with IBM watsonx.ai Embedding