NOTE: if on runpod remember to set HF_HOME to workspace cache (and HF_TOKEN too)
WARN: on run it will download "meta-llama/Meta-Llama-3-8B-Instruct" if its not present in current HF_HOME
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
I include https://github.com/vgel/repeng/
in this repo for the repeng
library
start the server and show the web interface
uvicorn main:app --reload
visit http://localhost:8000
or http://localhost:8000/docs
can load and test vectors or submit new ones to the training queue
python run_worker.py
worker will process the queue and train new vectors