Python Binding: Unnecessary Dependency on active internet connection with allow_download=True

[margau]

Bug Report

With allow_download=True, gpt4all needs an internet connection even if the model is already available.

Example Code

Steps to Reproduce

1. Start gpt4all with a python script (e.g. the example code) and allow_download=True (the default)
2. Let it download the model
3. Restart the script later while being offline
4. gpt4all crashes

Expected Behavior

It should work offline, when the model was downloaded already

Your Environment

• Bindings version (e.g. "Version" from pip show gpt4all): 2.7
• Operating System: Linux

Proposed Solution

The model check (https://github.com/nomic-ai/gpt4all/blob/54ed30937f717286083c7071e288393a3c49a769/gpt4all-bindings/python/gpt4all/gpt4all.py#L309) should be only done when the local model is unavailable (https://github.com/nomic-ai/gpt4all/blob/54ed30937f717286083c7071e288393a3c49a769/gpt4all-bindings/python/gpt4all/gpt4all.py#L338) to allow using cached models without internet connectivity.

[cosmic-snow]

This is not a bug.

The constructor parameter allow_download is not only for retrieving models, but also for the metadata in models3.json to populate session templates. This was previously mentioned in the old documentation and you can now find that explanation in the wiki.

Does that resolve the issue for you?

[margau]

Solved, looked into https://docs.gpt4all.io/gpt4all_python/ref.html#gpt4all.gpt4all.GPT4All.__init__ and did not see the behavior in the "allow_download" description there.

[cosmic-snow]

No worries, I had a hunch it just wasn't clear.