Sinaptik-AI / pandas-ai

Chat with your database (SQL, CSV, pandas, polars, mongodb, noSQL, etc). PandasAI makes data analysis conversational using LLMs (GPT 3.5 / 4, Anthropic, VertexAI) and RAG.
https://pandas-ai.com
Other
12.09k stars 1.14k forks source link

Add support for VertexAI models #309

Closed gventuri closed 1 year ago

gventuri commented 1 year ago

🚀 The feature

We should add the support for VertexAI models (i.e. PaLM 2).

Motivation, pitch

Given this is backed by Google, it will become increasingly important.

Alternatives

No response

Additional context

mspronesti commented 1 year ago

Do you plan to support all models or only the code ones?

gventuri commented 1 year ago

@mspronesti it's a great question! Maybe all the models, provided that the non-code ones are good enough at generating code? We should definitely benchmark the quality of each model. What's your thoughts about it @mspronesti?

amjadraza commented 1 year ago

Let me find some literature on this, however, code-bison should be the default considering Google much have done some benchmarks already.

amjadraza commented 1 year ago

I have one point, should we include this support as it is already supported through LangChain Vertex Model?

gventuri commented 1 year ago

@amjadraza great, totally agree.

Since langchain is an optional dependency, I think supporting the most widespread LLMs "natively" would be a much smoother process. The goal of PandasAI has always be being fast and easy to use, so any unnecessary configuration should be removed imo!

amjadraza commented 1 year ago

@amjadraza great, totally agree.

Since langchain is an optional dependency, I think supporting the most widespread LLMs "natively" would be a much smoother process. The goal of PandasAI has always be being fast and easy to use, so any unnecessary configuration should be removed imo!

Sure, I can work on this feature. Would as for help if needed.

gventuri commented 1 year ago

Would be awesome, assigning it to you! Thanks a lot @amjadraza!

amjadraza commented 1 year ago

@gventuri please see the PR against this issue.