How to Deploy DeepSeek LLM Locally on JD Cloud GPU with Ollama and Chatbox

1. JD Cloud GPU Host Preparation

DeepSeek models have varying hardware requirements; a dedicated GPU (e.g., NVIDIA RTX 3090 or equivalent) is recommended, and the CPU should support AVX2/AVX‑512 for optimal performance.

1.1 Create GPU Instance

Select billing mode, region, CPU architecture, Ubuntu 22.04 LTS image, and a Tesla P40 GPU. The following screenshots illustrate the process.

Configure instance type, storage, network, security group (allow port 11434), and set a password before confirming purchase.

After creation, verify the instance runs successfully.

1.2 Install GPU Driver

Connect via SSH and run the following commands on Ubuntu 22.04 to install the recommended NVIDIA driver (550) and reboot.

<code>root@deepseek-vm:~# apt update
root@deepseek-vm:~# ubuntu-drivers devices
# Identify the recommended driver, e.g., nvidia-driver-550
root@deepseek-vm:~# apt install nvidia-driver-550 -y
root@deepseek-vm:~# reboot
root@deepseek-vm:~# nvidia-smi   # verify driver and GPU status</code>

Sample

nvidia-smi

output shows two Tesla P40 GPUs with no memory usage.

2. Deploy Ollama

Ollama is an open‑source LLM service that simplifies local deployment.

2.1 Download Binary Package

Download the Linux amd64 tarball (v0.5.7) from the provided JD Cloud OSS link or from the official GitHub releases.

<code>root@deepseek-vm:~# cd /usr/local/src/
root@deepseek-vm:/usr/local/src# wget https://myserver.s3.cn-north-1.jdcloud-oss.com/ollama-linux-amd64.tgz</code>

2.2 Install and Run Ollama

<code>root@deepseek-vm:/usr/local/src# tar -C /usr -xzf ollama-linux-amd64.tgz
root@deepseek-vm:/usr/local/src# ollama serve   # start the service</code>

Ollama generates a new SSH key on first run and logs startup information, including the listening address

127.0.0.1:11434

.

2.3 Verify Ollama Is Running

<code>root@deepseek-vm:~# ollama -v
ollama version is 0.5.7</code>

2.4 Create Systemd Service

<code># /etc/systemd/system/ollama.service
[Unit]
Description=Ollama Service
After=network-online.target

[Service]
ExecStart=/usr/bin/ollama serve
Environment="OLLAMA_HOST=0.0.0.0:11434"
User=ollama
Group=ollama
Restart=always
RestartSec=3
Environment="PATH=$PATH"

[Install]
WantedBy=default.target</code>

Enable and start the service:

<code>root@deepseek-vm:~# systemctl daemon-reload
root@deepseek-vm:~# systemctl enable ollama
root@deepseek-vm:~# systemctl start ollama
root@deepseek-vm:~# systemctl status ollama</code>

3. Run DeepSeek Model via Ollama

Use Ollama to pull and run DeepSeek‑R1 models. The 1.5B model requires ~1.1 GB download; the 8B model requires ~4.9 GB.

<code># Run 1.5B model
root@deepseek-vm:~# ollama run deepseek-r1:1.5b

# Run 8B model
root@deepseek-vm:~# ollama run deepseek-r1:8b</code>

After download, an interactive CLI appears for chatting with the model.

4. Graphical Client – Chatbox

Chatbox provides a cross‑platform GUI for interacting with Ollama‑served models.

4.1 Windows Installation

Download the Windows installer, run it, choose installation scope and path, then launch Chatbox.

4.2 Configure Chatbox

In Settings, select "Ollama API", set the endpoint to

http://<your‑host>:11434

, and choose the desired DeepSeek model version.

4.3 Test Conversation

Ask sample questions such as "Are there aliens?" or "Explain Kubernetes"; Chatbox displays the model's responses.

4.4 macOS Installation

Download the macOS zip, extract, move the app to

/Applications

, and trust the developer in System Settings.

Configure the same Ollama endpoint as on Windows and test the conversation.

5. Local Data Feeding with AnythingLLM

AnythingLLM can ingest local documents and use DeepSeek as the embedding model, enabling a private knowledge base.

5.1 Install AnythingLLM

Download the Docker compose file or binary from the official repository and start the service.

5.2 Basic Configuration

Set the LLM preference to DeepSeek (via Ollama), choose "Query" mode, and configure the embedder to use Ollama.

5.3 Upload Documents

Upload PDFs, TXT, Word, Excel, or PPT files; the system extracts text and indexes it.

5.4 Query the Knowledge Base

Ask questions that are answered using the uploaded documents; if the query is outside the knowledge base, the model falls back to its general knowledge.

This guide provides a complete end‑to‑end workflow for deploying DeepSeek locally, exposing it via Ollama, interacting through a graphical client, and extending its capabilities with a private document‑based knowledge base.