Multimodal Tool Calling Pipeline

Implement built-in tool calling in open-sourced multimodal models to enable any-to-any conversions, like transcribing 30 minutes of audio within a single model interface.