Retrieval-Augmented Generation (RAG) on Palmetto HPC

Retrieval-Augmented Generation (RAG) on Palmetto HPC#

Instructor#

  • Instructor: Carl Ehrett

  • Office: 2105 Barre Hall, Clemson University

  • Email: cehrett AT clemson DOT edu

Workshop Description#

This one-day, hands-on workshop teaches Retrieval-Augmented Generation (RAG) for research and how to run it efficiently on Clemson’s Palmetto HPC cluster. We will cover core RAG concepts, semantic embeddings and retrieval, an end-to-end pipeline with citations, and a graph-based RAG extension. Participants leave with working code and clear patterns for scaling and adapting RAG to their own datasets on Palmetto.

Prerequisites#

Other setup needed#

In order to use the code in the Workshop notebooks, you will need a Hugging Face account. You can create one here. After doing so, please get an Access Token for your Hugging Face account (here). You should then connect your LLMsFT python env to your Hugging Face account by activating that python env in a terminal on the cluster, and running huggingface-cli login (i.e., in a terminal on the cluster, first run module load miniforge3 then source activate LLMsFT and finally huggingface-cli login). You will be prompted to supply your access token. It is recommended that you also add the line export HF_HOME=/scratch/[YOUR PALMETTO USERNAME]/hf_cache/ to your ~/.bashrc file, so that pre-trained LLMs will be downloaded to your scratch drive.