The community member got a code from YouTube that worked fine, but when they tried to use a sub question query, it caused an error. The comments suggest that the issue is related to the sub question query engine, which relies on the language model (LLM) being able to write JSON that can be parsed. However, open-source LLMs often struggle with this. The comments also mention that the query wrapper prompt is incorrect for the llama2-chat model, and it should be something like "<s> [INST] {query_str} [/INST] ". The community members suggest that to use the sub question query, the user should either use models from OpenAI or find other open-source models that can write JSON effectively. Alternatively, they could use OpenAI for the question generator and another LLM for the rest of the system.
