Running Your Fine-Tuned Model in Ollama

You have a trained adapter sitting in $HOME/companion/code/finetuning/granite_sql_lora/, but you cannot run it in Ollama yet. The adapter is only the small change you trained, not a complete model, and it is in a format meant for training tools, not for Ollama. This section turns it into a model Ollama can run.

There are 4 things to do, and the good news is that Unsloth collapses the first two into a single command:

1. Merge the adapter into the base model, so you have one complete model instead of a base plus a separate adapter.