🤖 LLM + ✉️ email = 🔥 llmail

A bunch of experiments for running local LLM-s to label incoming emails.

1. conda create -n llmail python=3.11
2. pip install -r requirements.txt
3. Create an app password for your gmail account
4. cp .env.exammple .env

You can try work in progress notebooks in the experiments folder.

I am looking to reduce llama 2 CPU latency as much as possible. Let me know if you have a good solution. I am exploring FHE/MPC, Speculative sampling and MoE ATM.