chatbot-ollama¶

• Main author - HE-Arc
• Code
• Deployment configuration
• Staging
• Production (not available yet)

Description¶

This service is a streamlit interface to interact with a large language model using the ollama API and langchain.

Ollama is an open source project that aims to provide a simple and easy to use app to download and use LLM models like llama2 or mistral locally.

On the Ollama official website, you can find the list of available models and the link to the github repository.

Before using this service, you need to vectorize a pdf document using the document-vectorizer service.

Environment variables¶

All environment variables are described in the .env file.

The environment variables can be overwritten during the CI/CD pipeline described in the workflow.yml GitHub workflow file.

Start the service locally with Python¶

In the chatbot-ollama directory, start the service with the following commands.

1
2
3
4
5
6
7
8

# Generate the virtual environment
python3 -m venv .venv

# Activate the virtual environment
source .venv/bin/activate

# Install the requirements
pip install /code[ui]

Start the application.

1
2

# Start the application
streamlit run app/app.py --server.port 80

Access the service app on http://localhost:80.

Start the service locally with minikube and the Docker image hosted on GitHub¶

Start the service with the following commands. This will start the service with the official Docker images that are hosted on GitHub.

In the chatbot-ollama directory, start the service with the following commands.

1
2
3
4
5

# Start the chatbot-ollama
kubectl apply \
    -f kubernetes/chatbot-ollama.config-map.yml \
    -f kubernetes/chatbot-ollama.stateful.yml \
    -f kubernetes/chatbot-ollama.service.yml

Create a tunnel to access the Kubernetes cluster from the local machine. The terminal in which the tunnel is created must stay open.

1
2

# Open a tunnel to the Kubernetes cluster
minikube tunnel --bind-address 127.0.0.1

Access the chatbot-ollama app on http://localhost:9090/.

Then you must use the document-vectorizer service to vectorize a pdf document before using the chatbot-ollama app. See the document-vectorizer service documentation to deploy it locally.

Start the service locally with minikube and a local Docker image¶

Note: The service StatefulSet (chatbot-ollama.stateful.yml file) must be deleted and recreated every time a new Docker image is created.

Start the service with the following commands. This will start the service with the a local Docker image for the service.

In the chatbot-ollama directory, build the Docker image with the following commands.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18

# Access the Minikube's Docker environment
eval $(minikube docker-env)

# Build the Docker image
docker build -t ghcr.io/swiss-ai-center/chatbot-ollama-service:latest .

# Exit the Minikube's Docker environment
eval $(minikube docker-env -u)

# Edit the `kubernetes/chatbot-ollama.stateful.yml` file to use the local image by uncommented the line `imagePullPolicy`
#
# From
#
#        # imagePullPolicy: Never
#
# To
#
#        imagePullPolicy: Never

In the chatbot-ollama directory, start the service with the following commands.

1
2
3
4
5

# Start the chatbot-ollama
kubectl apply \
    -f kubernetes/chatbot-ollama.config-map.yml \
    -f kubernetes/chatbot-ollama.stateful.yml \
    -f kubernetes/chatbot-ollama.service.yml

Create a tunnel to access the Kubernetes cluster from the local machine. The terminal in which the tunnel is created must stay open.

1
2

# Open a tunnel to the Kubernetes cluster
minikube tunnel --bind-address 127.0.0.1

Access the chatbot-ollama app on http://localhost:9090/docs.

Access the Core engine documentation on http://localhost:8080/docs to validate the backend has been successfully registered to the Core engine.