We’ve been working on a demo that showcases RAG (Retrieval-Augmented Generation) using open-source models directly within Postgres. It's a Wikipedia chatbot built with PostgresML.
If you haven’t seen us here before, PostgresML is an open-source extension for Postgres that lets you perform ML/AI inside PostgreSQL. There are a number of performance benefits that come with doing machine learning in-database. It’s how our team built and scaled the ML platform at Instacart during Covid, so it’s a thesis that’s been battle tested to say the least.
Key points:
* Fully open-source stack
* Performs the entire RAG workflow (LLMs, vector memory, embedding generation, re-ranking, summarization) in a single SQL query
* RAG performed in-database
We’re curious to hear your thoughts, especially from those who've worked with other RAG implementations or in-database ML. Any advantages/drawbacks you're seeing with this approach?
We built an open-source AI SDK (Python & JavaScript)that provides a drop-in replacement for OpenAI’s chat completion endpoint.
Let us know what you think so we can make switching as easy as possible and get more folks on open-source.
If you haven't seen us here before, we're PostgresML, an open-source MLOps platform built on Postgres. We bring ML to your data(base) rather than the other way around. We're incredibly passionate about keeping AI truly open. So we needed a way for our customers to easily move off of OpenAI's closed models. Especially now that there are better performing models, anyways.
You can check out the doc linked or the blog post from when we first released it, but here's the git diff:
- from openai import OpenAI + from pgml import OpenSource AI
We’ve been working on a demo that showcases RAG (Retrieval-Augmented Generation) using open-source models directly within Postgres. It's a Wikipedia chatbot built with PostgresML.
If you haven’t seen us here before, PostgresML is an open-source extension for Postgres that lets you perform ML/AI inside PostgreSQL. There are a number of performance benefits that come with doing machine learning in-database. It’s how our team built and scaled the ML platform at Instacart during Covid, so it’s a thesis that’s been battle tested to say the least.
Key points: * Fully open-source stack * Performs the entire RAG workflow (LLMs, vector memory, embedding generation, re-ranking, summarization) in a single SQL query * RAG performed in-database
We’re curious to hear your thoughts, especially from those who've worked with other RAG implementations or in-database ML. Any advantages/drawbacks you're seeing with this approach?
Try it out: https://postgresml.org/chatbot GitHub: https://github.com/postgresml/postgresml
Looking forward to your feedback and any questions about the technical details.