Closed erhant closed 2 weeks ago
Turns out ollama serve
itself accepts an OLLAMA_KEEP_ALIVE
variable, defaulting to 5m
which means 5 minutes. We should also take this from env.
Closed as if the node is active enough, this will not be a problem; if the model is active with multiple models we have to remove the model nevertheless.
Nodes may continue to store the LLM in their memory, otherwise memory will be freed by Ollama-rs after 5 minutes of inactivity.