Got "file upload" working with the latest Llama models.

Also, it seems the smallest Llama 1b (fine-tuned) model performs even better with more context.

Which is really good because that one is really fast even on a medium-spec device.