Remarks: I know there is a keep_alive option, but in my case I just want to close connection instead of unloading the model from memory

Hey! Can you point to your usage + what the OpenAI equivalent is? Thanks!

My usage

case 1 - using ollama python package

import ollama
response = ollama.chat(...)
...

case 1 results in warning on exit:

sys:1: ResourceWarning: unclosed <socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=6, laddr=('127.0.0.1', 34886), raddr=('127.0.0.1', 11434)>

case 2 - using openai equivalent

from openai import OpenAI
client = OpenAI(api_key="ollama", base_url="http://localhost:11434/v1")
client.chat.completions.create(...)

client.close()

In case 2, the final line client.close() saves me from the resource warning.

Could you try this and see if it helps - just as an experiment not recommending for actual usage:
client._client.close()

I tried your suggestion _client.close() with Ollama python library and got error:

AttributeError: 'Client' object has no attribute 'close'

Cool cool, thanks for testing. Will look into this and possibly expose something. Thanks for the report!

I am looking forward to the fix. Currently, I am working on a project that supports 14 AI backends, of which only Ollama and Cohere are experiencing connection issues, while the rest are functioning properly. Since I have set Ollama as the default backend, the issue is somewhat embarrassing.