Structured

The community members are discussing issues with using the Google Gemma model with the LlamaIndex library. They are encountering a "json" error when trying to use the SQLAutoVectorQueryEngine and SQLJoinQueryEngine examples. The community members suggest that features requiring structured outputs may not work well with open-source models, and that a fairly capable language model is generally required. They discuss whether fine-tuning the language model could be helpful, particularly for outputting JSON that LlamaIndex can parse. Some community members provide suggestions, such as enabling debug logs and using the llama_index.core.set_global_handler("simple") function. The community members also discuss their experience with the Mistral library and whether fine-tuning could be helpful in that case as well.