* add buffer cache limit * swift-format * a more reasonable size * add memory stats to command line tool, update to final api * add note about changing models
- runs on iOS and macOS - downloads a model / tokenizer from hugging face - evaluates the given prompt