How to best format a schema for pip-sql?
Hello, I'm having some difficulties sometimes getting joins between tables in the schema properly recognised sometimes.
Is there any documentation out there about how to best specify primary and foreign keys the schema?
From what I can see in the training data schemas those relationships aren't specified, and everything is either type integer or text.
Does that pattern need to be followed?
Thank you!
Hi , we have mixed different prompt formats while training so as to ensure certain level of independence to prompt format. We will be releasing a new model next week where more type of instructions can be specified in pre prompt.
Adding as comment beside the col in prompt or adding instructions helps in many places right now.
BTW if you would like to colab with us to tune the model on your schema with our RL algo we can help out as well. Usually works best in production deployment settings where it has seen some 30 -40 example queries of the schema during finetuning.
Hi , try https://huggingface.co/PipableAI/pip-library-etl-1.3b .
Its more general purpose and can take instructions better.
If it does something wrong you can add that example in prompt and it will handle it's permutation type questions well. --qesution: --response: