# For Node Runners

## **Introduction**

As a **Node Runner**, you contribute GPU-powered computation to the **DIVS decentralized network**, running **Vision-Language Models (VLMs)** to verify claims in images submitted by Builders.

Node Runners help build a **trustless, censorship-resistant truth layer for online images**.

This guide walks you through installing the node client, configuring your compute resources, and starting your first verification tasks.

***

## **1️⃣ Prerequisites**

✅ **Hardware Requirements:** We support a range of models — from GPU-hungry giants to ones that can chill on your laptop.

* **GPU (recommended):** NVIDIA with CUDA is ideal. More VRAM = happier models.
* **CPU (possible):** 4+ cores, but It’ll work… eventually. Great time to grab a coffee. Or two.
* **RAM:** At least 8GB. For bigger models, 16GB+ is safer.
* **Architecture**: x86\_64 and ARM supported (some models prefer x86 + CUDA).
* **Disk:** 120+ GB free space for keeping things comfy

✅ **Software Requirements**

* **Docker (v20+)** for containerized setup. That’s it. No additional installs or builds — just pull the image and run.

✅  **Network Requirements**

Our protocol uses peer-to-peer communication over UDP.

* **Ports:** Open **UDP ports 12000–12009** on your router or firewall.
* **Connectivity:** A stable public internet connection is best. NAT traversal is attempted, but port forwarding is recommended.
* **Docker note:** Make sure Docker can expose the above ports correctly.

✅ **DIVS Wallet Configuration**

* **Automatic:** Node keys are auto-generated on first run.
* **Optional override:** You can supply your own key using environment variables when starting the container.

***

## **2️⃣ Run the DIVS node**

### Create a Volume 

So that the node will not pull models again and again

```
docker volume create wtns-vol
```

### 🚀 Option 1: With NVIDIA GPU (Recommended) <a href="#option-1-with-nvidia-gpu-recommended" id="option-1-with-nvidia-gpu-recommended"></a>

For the best performance and support for larger models, run your Watchtower using a CUDA-enabled NVIDIA GPU:

```
docker run \
 -d \
 --gpus all \
 --network=host \
 -v wtns-vol:/root \
 -e WALLET_PUBLIC_KEY=0x_your_key_here \
 -e MODEL_NAME=MODEL_NAME \
 -e NETWORK=testnet \
 --name mywatchtower \
 witnesschain/infinity-watch-nvidia:2.0.0
```

### 🧪 Option 2: CPU-Only (Lightweight Model) <a href="#option-2-cpu-only-lightweight-model" id="option-2-cpu-only-lightweight-model"></a>

No GPU? You can still join the network by running a smaller model on your CPU:

```
docker run \
 -d \
 -v wtns-vol:/root \
 -e WALLET_PUBLIC_KEY=0x_your_key_here \
 -e MODEL_NAME=MODEL_NAME \
 -e NETWORK=testnet \
 --name mywatchtower \
 witnesschain/infinity-watch:2.0.0
```

{% hint style="info" %}
Note: if you want to use your own private key for your watchtower,  add the environment variable PRIVATE\_KEY in your docker run command

-e PRIVATE\_KEY="your\_custom\_private\_key"
{% endhint %}

### **Models Supported**

Following are the models we support as of now. Use the below models to pick one for the MODEL\_NAME variable.

{% hint style="success" %}
We keep adding models frequently. If you want to add your model to the list, write to us at <support@witnesschain.com>
{% endhint %}

<table><thead><tr><th width="479.91796875">🤗 HuggingFace family </th><th align="center">RAM Requirement</th></tr></thead><tbody><tr><td>HuggingFaceTB/SmolVLM2-2.2B-Instruct</td><td align="center">6 GB</td></tr><tr><td>HuggingFaceTB/SmolVLM-500M-Instruct</td><td align="center">2 GB</td></tr><tr><td>HuggingFaceTB/SmolVLM-256M-Instruct</td><td align="center">1 GB</td></tr></tbody></table>

<table><thead><tr><th width="479.5625">🔮 Qwen Family</th><th align="center">RAM Requirement</th></tr></thead><tbody><tr><td>Qwen/Qwen2.5-VL-7B-Instruct</td><td align="center">16 GB</td></tr><tr><td>Qwen/Qwen2.5-VL-3B-Instruct</td><td align="center">8 GB</td></tr></tbody></table>

<table><thead><tr><th width="480.06640625">🧠 GLM Family</th><th align="center">RAM Requirement</th></tr></thead><tbody><tr><td>zai-org/GLM-4.1V-9B-Thinking</td><td align="center">22 GB</td></tr></tbody></table>


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.witnesschain.com/infinity-watch/proof-of-model-testnet/for-node-runners.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.