Evo2 docker image step‐by‐step - danifilho/Evo2_BASF GitHub Wiki

Evo2 Container Setup & Test Workflow

This document describes how to pull the Evo2 Singularity image, set up HuggingFace authentication, and run the provided test script inside the container from first person (me) POV.

---

Prerequisites

• Singularity (singularity-ce version 4.1.2-jammy)
• A HuggingFace account and an active access token with read permissions
• Python 3.10.14
• GPU drivers available in my case, running it on dev-amd24-h200 at HPCC (--nv flag)

---

1. Clonning the Evo2 repo

git clone --recurse-submodules https://github.com/ArcInstitute/evo2.git && cd evo2/

2. Pulling the Evo2 Singularity Image

In my project directory (where i’ll keep evo2_latest.sif):

singularity pull docker://tezavortix/evo2

Running this in my project directory to download evo2_latest.sif.

---

3. Installing and logging at the HuggingFace Hub CLI

pip install --user --upgrade huggingface-hub
~/.local/bin/huggingface-cli login

Following the prompts to log in, generating/updating ~/.cache/huggingface/token.

---

5. Preparing the Local HuggingFace Cache

mkdir -p $PWD/huggingface
cp ~/.cache/huggingface/token $PWD/huggingface/
chmod 600 $PWD/huggingface/token

---

6. Running the Test Script Inside the Container

singularity exec --nv \
  -B "$PWD/huggingface:/root/.cache/huggingface" \
  evo2_latest.sif \
  python3 ./test/test_evo2.py --model_name evo2_7b

I'm using --nv for GPU support and mount my local HF cache so the container can authenticate properly.