Faster Failure for ChatOllama object (at initialization)

[DanielYakubov]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Minimal example:

from langchain_ollama import ChatOllama

llm = ChatOllama(model="fake_model")
result = llm.invoke(messages)

The current design fails at call time (the invoke line) rather than when the ChatOllama instance is initialized. I would guess this is the design throughout.

Motivation

IMO, it makes no sense to successfully initialize an object if it won't work. This would be better with a fail-fast philosophy.

Further motivation would be if a ChatOllama instance is initialized as an attribute of an object, but not used until a later flow, it may be missed in the QA of some applications.

Proposal (If applicable)

Model objects should fail to initialize if the model isn't pulled, this would require some refactoring of the code. I am more than happy to work on it myself, I just need to understand if the authors have a reason for this design