Adding a keep_alive parameter for Ollama models

[ashleyp33]

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

Ollama provides a parameter when making a curl request called keep_alive that takes a number in seconds or any negative number to keep it alive forever. This parameter should be added when defining an Ollama llm.

Motivation

This will make sure the model stays loaded for the amount of time specified and speeds up concurrent calls. There doesn't seem to be a way to set this on the server and is done only through the request, so it would be nice to customize that when using Ollama with langchain. It does seem like there is a keep alive parameter built into the request already as the model does not offload immediately, but specifying a time would be nice for production applications.

Proposal (If applicable)

No response