OLLAMA Api Base not considered

[octera]

Hello,
I'm trying to wire pr-agent with a local LLM running with OLLAMA.
I have followed the docs with settings the ollama api base in the secrets file or throught env var : None is considered by the PR Agent.
I think this came from this file pr_agent/algo/ai_handlers/litellm_ai_handler.py which is not considering this configuration : it keep the default URL : localhost:11434

I have been able to make a workaround with the huggingface configuration.

Exemple :
Not working :

CONFIG.MODEL=ollama/aRandomModel 
OLLAMA.API_BASE=http://ollama-service:11434

Working :

CONFIG.MODEL=ollama/aRandomModel 
HUGGINGFACE.KEY="dummy" 
HUGGINGFACE.API_BASE=http://ollama-service:11434 

[hussam789]

/help

[gregoryboue]

Hi,

@octera your workaround is working until v0.12.
It's not working anymore since this commit with the following conditions :

if get_settings().get("HUGGINGFACE.API_BASE", None) and 'huggingface' in get_settings().config.model

So if there isn't 'huggingface' in the model name, it won't set api_base configuration.