This dataset contains pairs of natural language instructions (simulating commands given to Google Street View) and their corresponding structured JSON outputs representing the intended navigation action. It was generated using the Gemini API (gemini-1.5-flash-latest) based on predefined templates and few-shot examples.
The primary intended use is for fine-tuning small language models (like TinyLlama) to act as a translation layer between… See the full description on the dataset page:
https://huggingface.co/datasets/cahlen/streetview-commands-dataset.
