Integration with LangChain?

[bborn]

Hi! This is such a cool project - wondering if you've thought at all about integrating with LangChain or writing an adapter? Is there any reason why this has to be used with BERT? Or could it also work with other models?

[abhijithneilabraham]

@bborn I am pretty sure the scope of this project can go beyond using BERT-based models. What are your thoughts about integrating with LangChain? Would love to hear about it.

Best Regards.

[bborn]

LC has a SQLDatabase chain (https://langchain.readthedocs.io/en/latest/modules/chains/examples/sqlite.html?highlight=sql) for interacting with a database in natural language, but I think this repo is quite a bit more advanced.

I think it would be cool to make tableQA available as a LangChain tool (or Chain, maybe). Seems like the main thing that would need to change is making tableQA's LLM pluggable, instead of hard coding it to use Bert.

I don't know your codebase, but maybe starting here? https://github.com/abhijithneilabraham/tableQA/blob/master/tableqa/nlp.py#L12

Maybe that could be configurable so you could just plug any llm there (see here: https://langchain.readthedocs.io/en/latest/modules/llms/getting_started.html)

(Sorry if I'm off base here)

[abhijithneilabraham]

This makes a lot of sense to me! Do you think I would need extra compute for writing tests for this one? @bborn

[bborn]

Not sure - but I think you could use this: https://langchain.readthedocs.io/en/latest/modules/llms/examples/fake_llm.html?highlight=fake%20llm to avoid racking up $$

[abhijithneilabraham]

Thank You very much! I am definitely working on this if my work could be actually of use anywhere haha!