Step-by-Step Guidance

Welcome to the Step-by-Step Guidance version of this project. Let's do this!

📣 If you're EVER stuck - ask the NextWork community. Students like you are already asking questions about this project.

Before we start Step #1...

Use DeepSeek On Your Browser

We’ll start with the easiest way to use DeepSeek - its web app.

If you’ve used ChatGPT or any other LLM in your browser, this will feel familiar. But, as we dive into more advanced prompts, you’ll start noticing what makes DeepSeek stand out.

In this step, you're going to:

Create a DeepSeek account.
Run some fun prompts and see what makes DeepSeek stand out.
Spot some limitations of DeepSeek on your browser.

Create a DeepSeek Account

Open your web browser and head to DeepSeek's website.

Enter your details, like your email address and password.

🙋‍♀️ I don't want to enter my email
You could consider using a temporary email address instead.

For example, you could use a tool like TempMail to set up a temporary email address that receives mail for 1-2 hours, and deletes itself right after. Just make sure to keep this tab open until the end of the project, so you don't lose the inbox.

Select Send code.

Check your inbox - you should get a code sent to your inbox.

🙋‍♀️ I'm not getting a code
You might not get a code if DeepSeek's registration is busy. You could try signing up with Google instead. If Google isn't an option or isn't working either, you could skip this step for now and head to the next step. The first step is great for easy access to DeepSeek, but you can always come back to it later!

Enter the code in the #Code section of the sign-up form.
Select Sign Up.

Run a Prompt

Once you're logged in, enter your first prompt:

Summarize the fall of the Roman Empire using only text abbreviations and emojis.

💡 Tip: You can choose another prompt to give DeepSeek. We'd recommend a short and easy request (e.g. ask for a 100 word summary instead of a 1000 word essay), so you won't be waiting too long for an answer!

Select Send (or press Enter on your keyboard).
Nice! DeepSeek gives us an answer right away.

If you've used other AI-powered chatbot like ChatGPT or Claude, DeepSeek might feel pretty familiar. Great! That's because the web app uses DeepSeek's V-3 model by default, which is not DeepSeek's latest model, called R1. Let's try using R1 now.
Select the DeepThink (R1) button in your chat.

Enter the same prompt again. You can stay in the same chat window.

Summarize the fall of the Roman Empire using only text abbreviations and emojis.

Woah! This time, you can see DeepSeek's thinking process before it writes an answer.

💡 What is DeepThink (R1)?
DeepThink (R1) is DeepSeek's latest AI model. It stands out for displaying its real-time reasoning process before generating a response.

Reasoning is a big deal in LLMs. Unlike older models that simply predict the next word in a sequence, DeepThink - as well as OpenAI’s o1 model - is one of the first models to actively reviews its own responses as it generates them.

In R1's thinking process, spot for intermediate thinking steps like self-doubt ("hmm") and verification checks ("wait"), which gives you a lot of transparency into its problem-solving approach. Reasoning also makes an LLM much more efficient - they're more likely to solve a problem in one go, without requiring lots of back and forth between you and the LLM.

Test DeepSeek vs ChatGPT on Advanced Reasoning

Let's challenge DeepSeek to a harder prompt that requires more reasoning, and see how it compares with another LLM (e.g. OpenAI).

Self Host DeepSeek

Now that you have a feel of how DeepSeek works, let's see how we can host it locally without relying on the web app.

💡 What are the downsides of using a web app?
Ooo good question! Web-based LLMs, like ChatGPT and DeepSeek online:

Require constant internet connection (you can't go offline)

Introduce latency (web apps run slower when there are lots of people sending requests)

Process queries through external servers - which might raise privacy concerns around how the data is stored and used.

💡 What does it mean to host DeepSeek locally?
Running DeepSeek on your own computer means you don’t need the web app at all. Your device does all the processing, so no external servers are involved.

That means you can use DeepSeek offline, keep all your data private, and get faster responses since there's no waiting on the internet.

In this step, you're going to:

Install a tool (called Ollama) that lets you host DeepSeek on your computer.

Download Ollama

Note: If you already have Ollama installed, you can skip ahead to the next step.

Visit Ollama.

💡 What is Ollama?
Ollama is a tool that makes it easy to host LLMs, like DeepSeek, on your own computer. You can start chatting with LLMs over your computer's terminal!

Ollama takes care of downloading, installing, and running the models, so you don't have to worry about the complex setup that comes with hosting an LLM locally.

Ollama also gives you more control around the LLM you're using. We'll experiment with a setting called temperature later in this project to see the benefits of having wider control.

Select the Download button at the centre of the page.
Select your operating system.

Select the Download option for your operating system.
While we're waiting for Ollama to download...

Install Ollama

Next up, installing Ollama! Installation instructions depend on your operating system.

💡 Haven't I already installed Ollama?
So far you've just downloaded Ollama's installation files, which means Ollama is like a package that's been delivered to your door - but you haven't opened the package yet.

You'll need to open the package and set up permissions to start using Ollama's software in your computer.

Open your Mac's Downloads folder.
Select the Ollama zip file.

Open the Ollama app.

Select Move to Applications.

Nice! Now Ollama will take you through the process of installing the software locally.

Select Install.

Enter the password you use to unlock your Mac. This gives Ollama the permission to install the software into your computer.
In the next panel, we'll ignore the command that Ollama gives us. The command lets us use another open-source LLM, but we'll go straight to using DeepSeek.
Open your Terminal

After downloading the executable file, simply run it, and Ollama will be installed automatically.
If you're stuck or need any troubleshooting, we'd recommend checking out the Windows instructions on Ollama's GitHub.

For the most up to date instructions, we'd recommend visiting Ollama's GitHub

To install Ollama, run the following command:

curl -fsSL https://ollama.com/install.sh | sh

Manual install

[!NOTE] If you are upgrading from a prior version, you should remove the old libraries with sudo rm -rf /usr/lib/ollama first.

Download and extract the package:

curl -L https://ollama.com/download/ollama-linux-amd64.tgz -o ollama-linux-amd64.tgz sudo tar -C /usr -xzf ollama-linux-amd64.tgz

Start Ollama:

ollama serve

In another terminal, verify that Ollama is running:

ollama -v

AMD GPU install

If you have an AMD GPU, also download and extract the additional ROCm package:

curl -L https://ollama.com/download/ollama-linux-amd64-rocm.tgz -o ollama-linux-amd64-rocm.tgz sudo tar -C /usr -xzf ollama-linux-amd64-rocm.tgz

ARM64 install

Download and extract the ARM64-specific package:

curl -L https://ollama.com/download/ollama-linux-arm64.tgz -o ollama-linux-arm64.tgz sudo tar -C /usr -xzf ollama-linux-arm64.tgz

Adding Ollama as a startup service (recommended)

Create a user and group for Ollama:

sudo useradd -r -s /bin/false -U -m -d /usr/share/ollama ollama sudo usermod -a -G ollama $(whoami)

Create a service file in /etc/systemd/system/ollama.service:

[Unit] Description=Ollama Service After=network-online.target [Service] ExecStart=/usr/bin/ollama serve User=ollama Group=ollama Restart=always RestartSec=3 Environment="PATH=$PATH" [Install] WantedBy=default.target

Then start the service:

sudo systemctl daemon-reload sudo systemctl enable ollama

Run the following command to confirm you've installed Ollama correctly:

ollama --version

If Ollama is installed correctly, this command will print the version of Ollama you installed. Great success!

Access DeepSeek in the Terminal

Now that we've installed Ollama, how do we use it to access DeepSeek locally?

In this step, you're going to:

Find different DeepSeek models in Ollama.
Use DeepSeek in your local terminal.
Run some test prompts.

Find and Install DeepSeek R1

In a new tab, head to the Ollama library.

Before we look for DeepSeek, search for OpenAI

💡 Why can't I find OpenAI's models on Ollama?
Ollama focuses on open-source models like DeepSeek.

OpenAI's models are closed systems, so the underlying architecture, codebase, and datasets used to develop OpenAI models are confidential. Because they're confidential, it's not possible to use OpenAI locally in your machine.

Search for and select deepseek-r1

Select the dropdown labelled 1.5b.

💡 What are these different dropdown options?

The different dropdown options represent different model sizes for R1. Think of DeepSeek R1 in the web app as R1 at full capacity - if you wanted to run this version of R1 locally, you would need a computer with very large processing power and storage space (the dropdown tells us it requires 404GB of storage). This is far beyond what most computers can handle, as computers typically have less available storage and memory.

Model sizes let you choose a smaller, more accessible version of DeepSeek R1 for local use.

Smaller models (like 1.5b) are faster and require less memory to run locally, while larger models (like 8b) have deeper reasoning abilities and are more accurate. We're installing 1.5b first as a quick start, but we'll use a larger model next to the difference in performance.

💡 What does "1.5b" mean?
In AI models like DeepSeek, "1.5 b" means the model has 1.5 billion parameters to learn patterns from data.

Think of parameters as tiny decision-makers inside the model, each helping it recognize patterns, analyze data, and improve reasoning. More parameters generally mean the model can handle more complex tasks, but bigger isn’t always better - it also depends on how well the model is trained.

Copy the Ollama command to the right of the model dropdown.

💡 What does this command do?
This command sets up DeepSeek's smallest model, i.e. the 1.5 billion parameter model, locally in your computer. Because the command uses run, your terminal will transform into a chat session with DeepSeek R1 too.

Head back to your computer's terminal.

Paste the code in a new line, and select Enter on your keyboard.

Ollama will take a minute to install 1.5b.
While we wait...

💡 Extra for Experts: The terminal response starts by 'pulling manifest' - what does that mean?

When you run the Ollama command, it fetches the DeepSeek model's manifest, which is like a blueprint that tells your computer how to set up and run the model. It includes instructions for downloading and configuring everything correctly.

The actual brain of DeepSeek is the model itself, which gets downloaded after the manifest. Think of the manifest as the setup guide, while the model is the intelligence your computer will use to process prompts and generate responses.

When you're ready, you'll see a prompt that tells you to send a message.
Say less! Enter your first prompt for DeepSeek over the terminal.
Enter Hello

Press Enter on your keyboard.
Woohooo! Well done - DeepSeek has just responded to your request.

Enter /bye in the terminal. This will end your session with DeepSeek.

Use Another DeepSeek R1 Model

Head back to the Ollama tab on DeepSeek R1.
Select another model that your computer can run with its storage capacity.

🙋‍♀️ How do I know how much storage my computer has?

Click the Apple logo in the top-left corner of your screen.
Select About This Mac.
Navigate to the Storage tab.
You'll see a breakdown of your storage usage and available space.

Option 1: Settings
- Click on the Start menu and select Settings (the gear icon).
- Navigate to System > Storage.
- You'll see a breakdown of your storage usage and available space.
Option 2: File Explorer
- Open File Explorer.
- Click on This PC in the left sidebar.
- Under Devices and drives, you'll see your storage devices with their available space.

Option 1: Terminal Command
- Open a terminal window.
- Type the following command and press Enter: df -h
- This will display a list of your file systems along with their used and available space.
Option 2: Disk Usage Analyzer
- Use the Disk Usage Analyzer tool (also known as Baobab).
- If it's not installed, you can install it using your package manager.
- Once opened, it provides a graphical representation of your disk usage.

If you're stuck picking a model size, we'd recommend going for the 8b option. If there are any issues with using it, you can always switch to the 7b option instead.

Back in your terminal, run the command that starts your larger DeepSeek model.

Test Prompts

This time, let's try a more complex prompt and do this offline.
Turn your computer's Wifi off - and remember to turn it back on after running the prompt below!
Type Can I use DeepSeek while offline? and press Enter on your keyboard.

Nice, you'll see that DeepSeek can still craft a response - even when Wifi is off!

💡 Why can I still access DeepSeek while offline?
Local hosting through Ollama means you don't need another server to process your prompt. The DeepSeek model is running entirely on your device without needing internet connectivity.

💡 What are the <think> tags?
The <think> tags are a terminal version of DeepSeek's real-time reasoning display, so you can still see how DeepSeek is generating its response.

You might've noticed that the the think tags were empty in your previous request. That's because Hello was a more straightforward prompt, so deep thinking (which triggers this real-time reasoning display) wasn't required.

Enter /bye in the terminal. This will end your session with DeepSeek.
Turn Wifi back on.

Use Chatbox with Ollama

While the terminal is great for quick tests, you might miss the look of the web app. It does a much better job of organizing your chats and making conversations user friendly!

No worries, you can use a tool called Chatbox to organize your conversations in the terminal to look like the web app too. Let's set that up!

In this step, you're going to:

Install Chatbox.
Connect Chatbox to Ollama.
Chat with both DeepSeek models.

Install and Configure Chatbox

Visit the Chatbox home page.

Select Download.

Select your operating system.

🙋‍♀️ How do I find the correct option for my operating system?

🍎 MacOS: Select the Apple icon from the top left hand corner of your computer's menu bar. Select About this Mac, and note whether your Chip says Apple (pick Apple Silicon) or Intel.

🖼️ Windows: Select the Start button and search for System Information. Note whether your System Type says x64-based or ARM-based PC.

🐧 Linux: Open a terminal and run uname -m. Match the output to the correct Linux package option.

Open Chatbox in the Downloads folder.

Add Chatbox to your Applications folder.

Open the Chatbox app.
Open your Chatbox Settings in the bottom left hand corner.

Select Ollama API as the Model provider.
Keep the default API Host.

💡 What does the Model provider setting do?
In Chatbox, the Model provider determines the API that will connect you to the LLM model you want to use. We're using the Ollama API, since Ollama is the tool we're using to run DeepSeek locally.

💡 Why are we leaving the API Host as the default?
To connect you with your local LLM, Ollama needs to set up an endpoint, which is like an address within your computer to run DeepSeek. Ollama sets up LLMs at a default location (127.0.0.1:11434), so we'll keep the default value in Chatbox. This setup lets Chatbox communicate directly with your locally-hosted DeepSeek model.

For the Model setting, choose your lighter model i.e. deepseek-r1:1.5b to start with.

Select Save.

Chat with DeepSeek

Select New Chat in the bottom left corner of Chatbox.

Send the prompt: How many r's are in strawberry?

While we're waiting for DeepSeek's response...

Read DeepSeek's response, does everything look correct?

🙋‍♀️ DeepSeek made an error!
You might notice that the 1.5b model gave you an incorrect answer! Instead of three r's in strawberry, it only found two.

The 1.5b model is the most lightweight R1 model, so it's less able to analyse text and conduct proper reasoning.

🙋‍♀️ DeepSeek didn't make an error!
How good is that! It's great if DeepSeek got that right with a smaller model. A comparison where both models produce the right result is still a great experiment. You could always try other problems or prompts to test the limits 🔥

Let's compare this with your larger local DeepSeek model!
Select another New Chat in the bottom left corner of Chatbox.
This time, let's switch the DeepSeek model from 1.5b to 8b.

Send the same prompt: How many r's are in strawberry?
Notice that your DeepSeek 8b model might take longer to conduct a deeper chain of reasoning. But, you might also get a more though-out answer!

Temperature Test

Let's explore an advanced setting, called temperature, to see how you can customize DeepSeek R1 depending on the use case.

In this step, you're going to:

Run the same prompt with different temperature settings in Chatbox.

High Temperature Test

In Chatbox, select your Settings from the bottom left hand corner again.

For the Model setting, switch to your heavier, more advanced model i.e. deepseek-r1:8b.
At the bottom of the panel, dial the Temperature setting up to 2

💡 What is temperature?
Temperature controls the randomness of an LLM's output.

A higher temperature, like 2, gives you more creative and unpredictable responses, while a lower temperature, like 0, gives more focused and logical responses.

This is a detailed setting that you might not have access to over a web app, but possible with local hosting and APIs. Chatbox makes it easy for you to edit and customise your AI model's temperature.

Select Save.
Let's put our higher temperature to the test!
Start a New Chat in Chatbox.
Input the following prompt:

Create a recipe for a dessert that includes avocados, chocolate, and sea salt.

While we wait for a response...

Review DeepSeek's response - can you spot some quirky phrases, ingredients or instructions in the response? 🕵️‍♀️

In this example, you might notice that the recipe uses orange juice! Interesting ingredient choice...

Select your Settings from the bottom left hand corner again.
Change the temperature to 0 and select Save.

Low Temperature Test

Let's put our lower temperature to the test!
Start a New Chat in Chatbox.
Input the same prompt:

Create a recipe for a dessert that includes avocados, chocolate, and sea salt.

While we wait for DeepSeek to reason and write a response, head back into your browser.

Set up a Third Chat

Open ChatGPT or your favourite LLM (e.g. Gemini, Claude, maybe even DeepSeek) in your browser.

💡 Why are we opening ChatGPT?
This new chat will act as the judge of the two responses we generate - can another AI tell the difference between high and low temperature responses?

You'll also get to learn a breakdown of how to detect low vs high temperature text along the way.

Enter a prompt to ChatGPT to make it your temperature test judge.

You are an AI master. I will give you two pieces of generated text that received the same prompt. One was generated with a high temperature, the other was generated with a low temperature. You are to identify which one was generated with a higher temperature setting.

Great! Let's tell ChatGPT that the first response (your high temperature test results) is ready now, but not the second:

The second response is still generating, but I have the first response ready for you now. Can you read the first response, then wait for the second response after?

Head back into Chatbox.
Copy the first response.

Send the first response to ChatGPT.

Now head back to the DeepSeek chat where your second respoonse has been generating.
Your second response might take more time to generate, so you might need to wait a few more minutes.
Review DeepSeek's response - do you see any difference in the tone of voice and choice of words? 🕵️‍♀️

Let's see if our judge can tell the difference.
Copy the second response.
Head back into ChatGPT.
Send the second response to ChatGPT in the same window.
Let's see the analysis!

Oooo, nice work ChatGPT. You might notice that ChatGPT can correctly point out the first response had a higher temperature setting, and why.

💡 Why does temperature matter?
Different temperature settings work great for different scenarios.

Translation services and code commenting usually use a low temperature setting (0.2-0.4) to prioritize accuracy.

Chatbots usually use a low-mid temperature setting (0.5), so that responses are true and accurate.

For summarising meetings or writing email responses, you might opt for a mid-high temperature setting (0.6-0.8) to add a bit of variety and generated insights.

For creative writing and brainstorming, opt for a higher temperature setting (0.9-1.5) to get a wide range of results.

💡 Extra for Experts: I want to try another temperature experiment! We got you! Challenge your DeepSeek model to generate two responses - one with high temperature (2), one with low temperature (0) - to the following prompt:
Write a short 100 word story set in a world where gravity changes direction every day.
See the differences between a creative high-temperature story, versus a logical low-temperature story!

The Token Efficiency Showdown

Welcome to your 🤫 exclusive 🤫 secret mission! Are you ready for the ultimate test?

Your mission, should you choose to accept it, is to expose how efficiently DeepSeek and OpenAI use tokens. This lets you know which model gives you the most value. Let’s dive in!

In this secret mission, you're going to:

Set up API access to OpenAI (for free).
Compare DeepSeek and OpenAI's efficiency at the same prompts.
Showcase advanced understanding of LLMs in your documentation!

Clean Up

Now that we've explored the world of LLMs with DeepSeek and Ollama, it's time to clean up. This is important to keep your systems tidy.

Resources to delete:

Remove the DeepSeek models from Ollama (optional).

Uninstall Ollama (optional).

Uninstall Chatbox (optional).

If you no longer plan to use the DeepSeek models, you can remove them to free up disk space.

In your terminal, run the following commands to remove the DeepSeek models:

ollama rm deepseek-r1:1.5b ollama rm deepseek-r1:8b

Run ollama list to see what models are left in your computer.
You should see an empty table!

In your terminal, run the following commands to stop and remove Ollama. You'll also remove any files Ollama has stored in your system:

🍎 MacOS

ollama stop sudo rm -rf /usr/local/bin/ollama sudo rm -rf ~/.ollama sudo rm -rf /Library/LaunchDaemons/com.ollama.ollama.plist sudo launchctl bootout system /Library/LaunchDaemons/com.ollama.ollama.plist

If you've installed Ollama via Homebrew, you can also uninstall it using

brew uninstall ollama brew cleanup

To check if Ollama is fully removed, run:

which ollama

If it returns ollama not found, Ollama is completely uninstalled.

🖼️ Windows

The Ollama Windows installer registers an Uninstaller application.

Under Add or remove programs in Windows Settings, you can uninstall Ollama.

🐧 Linux

Remove the Ollama service:

sudo systemctl stop ollama sudo systemctl disable ollama sudo rm /etc/systemd/system/ollama.service

Remove the Ollama binary from your bin directory (either /usr/local/bin, /usr/bin, or /bin):

sudo rm $(which ollama)

Remove the downloaded models and Ollama service user and group:

sudo rm -r /usr/share/ollama sudo userdel ollama sudo groupdel ollama

To uninstall Chatbox, simply drag the app from your Applications folder and into your computer's Trash.

That's a wrap!

You've journeyed into the fascinating world of LLMs, and emerged victorious! 🏆

You've learned how to:

🐋 Use DeepSeek both online and locally.
🌡️ Control the creativity of DeepSeek with temperature settings.
💎 Thoroughly compare DeepSeek vs OpenAI.

🚀 p.s. Does it say "Still tasks to complete!" at the bottom of the screen?

This means you still have screenshots left to upload, or questions left to answer!

Press Ctrl+F (Windows) or Command+F (Mac) on your keyboard.

Search for the text Return to later.

Jump straight to your incomplete tasks!

🙋‍♀️ Still stuck? Ask the community!