Why does GPT4AllEmbeddings need internet access?

[angry-garden-salad]

I'm trying to follow the rag example at https://github.com/langchain-ai/langgraph/blob/main/examples/rag/langgraph_rag_agent_llama3_local.ipynb. I was able to get it all working on my Mac, but after I continued my testing in an air gapped environment, it hung on the "Index" section. I moved back to an environment that had internet access and it started working again. I then turned off my wifi and the procedure below produced the following error:

# Add to vectorDB
vectorstore = Chroma.from_documents(
    documents=doc_splits,
    collection_name="rag-chroma",
    embedding=GPT4AllEmbeddings(),
)

Error: HTTPSConnectionPool(host='gpt4all.io', port=443): Max retries exceeded with url: [/models/models3.json](https://file+.vscode-resource.vscode-cdn.net/models/models3.json) (Caused by NameResolutionError("<urllib3.connection.HTTPSConnection object at 0x14ad74df0>: Failed to resolve 'gpt4all.io' ([Errno 8] nodename nor servname provided, or not known)"))

Is there a different embedding I can use that does not reach back out to the internet?

[stephenhky]

It happened to me too. I downloaded the model, and I test that it works by pointing to the location of the downloaded model. But it still needs Internet access....

[Luca-Pozzi]

Hi! I might be too late to help the OP, but I hope my response can be useful for future users coming across this discussion.
Even in the GPT4All Python SDK you have to explicitly pass the allow_download=False argument to prevent the object to access gpt4all.io in order to get the list of available models.

When using the GPT4All embeddings via LangChain, you have to pass allow_dowload as key-value pair of the gpt4all_kwargs dictionary, i.e.

from langchain_community.embeddings import GPT4AllEmbeddings
    
model_path = "path\\to\\models\\nomic-embed-text-v1.5.f16.gguf"
model = GPT4AllEmbeddings(model_name=model_path,
                          gpt4all_kwargs= {'allow_download':False}
                          )
embeddings = model.embed_documents[['This is a test string to embed,',
                                    'but I don\'t care about the result',
                                    'I just want to use my model offline.'
                                    ]]

It might be confusing, but it is the intended behavior of the parameter, as you can see in the GPT4All GitHub repo.