Browser-based AI Chatbot Served From The IC

General
32

Hi @devvspaces ,

Training a custom LLM using the PDFs is something you could look into. That capability is not build into DeVinci right now.

The IC is not yet able to support on-chain training, so you would need to do that off-chain, and then integrate the specially fine-tuned LLM into the dApp for Inference purposes.

What you are able to do in DeVinci today, is to attach the PDFs to a chat:

  • Go to https://k4mni-yaaaa-aaaal-qklla-cai.icp0.io/
  • Connect. (Login with Internet Identity)
  • Select a model (Eg. Gemma)
  • Wait until the download into your local browser is done
  • Now the Chat Window activates, and you can attach a PDF by clicking on the paper clip icon

image

That activates a RAG based solution, fully inside the browser.

One of the idea for the hackathon is that you try to expand on this approach of using a RAG approach to work towards a tutor concept.

1 Like