The post discusses the recent capabilities of Ollama for generating structured outputs from input prompts, particularly in relation to producing clean CSV data. Users express excitement over the ability to define specific output formats without extraneous text, thus streamlining data extraction processes. Experiences with various models, including gemma2:2b, show promising results in generating structured JSON outputs. Community members seek insights regarding performance on AMD GPUs, the impact of constraints on output quality, and the best methods for structuring inputs effectively. There is a noticeable demand for enhancements in user support and clear communication from the maintainers of the tool, alongside discussions about alternative tools like oobabooga which provide efficient inference solutions. The overarching sentiment suggests that while Ollama presents significant improvements, users are navigating challenges related to model compatibility and the need for more robust documentation and support.