🧱 BrickLLM

BrickLLM is a Python library for generating RDF files following the BrickSchema ontology using Large Language Models (LLMs).

🧰 Features

• Generate BrickSchema-compliant RDF files from natural language descriptions of buildings and facilities
• Support for multiple LLM providers (OpenAI, Anthropic, Fireworks)
• Customizable graph execution with LangGraph
• Easy-to-use API for integrating with existing projects

💻 Installation

You can install BrickLLM using pip:

pip install brickllm

Development Installation

Poetry is used for dependency management during development. To install BrickLLM for contributing, follow these steps:

# Clone the repository
git clone https://github.com/EURAC-EEBgroup/brickllm-lib.git
cd brick-llm

# Create a virtual environment
python -m venv .venv

# Activate the virtual environment
source .venv/bin/activate # Linux/Mac
.venv\Scripts\activate # Windows

# Install Poetry and dependencies
pip install poetry
poetry install

# Install pre-commit hooks
pre-commit install

🚀 Quick Start

Here's a simple example of how to use BrickLLM:

Note

You must first create a .env file with the API keys of the specified LLM provider (if not local) and load them in the environment

from brickllm.graphs import BrickSchemaGraph

building_description = """
I have a building located in Bolzano.
It has 3 floors and each floor has 1 office.
There are 2 rooms in each office and each room has three sensors:
- Temperature sensor;
- Humidity sensor;
- CO sensor.
"""

# Create an instance of BrickSchemaGraph with a predefined provider
brick_graph = BrickSchemaGraph(model="openai")

# Display the graph structure
brick_graph.display()

# Prepare input data
input_data = {
    "user_prompt": building_description
}

# Run the graph
result = brick_graph.run(input_data=input_data, stream=False)

# Print the result
print(result)

# save the result to a file
brick_graph.save_ttl_output("my_building.ttl")

Using Custom LLM Models

BrickLLM supports using custom LLM models. Here's an example using OpenAI's GPT-4o:

from brickllm.graphs import BrickSchemaGraph
from langchain_openai import ChatOpenAI

custom_model = ChatOpenAI(temperature=0, model="gpt-4o")
brick_graph = BrickSchemaGraph(model=custom_model)

# Prepare input data
input_data = {
    "user_prompt": building_description
}

# Run the graph with the custom model
result = brick_graph.run(input_data=input_data, stream=False)

Using Local LLM Models

BrickLLM supports using local LLM models employing the Ollama framework. Currently, only our finetuned model is supported.

Option 1: Using Docker Compose

You can easily set up and run the Ollama environment using Docker Compose. The finetuned model file will be automatically downloaded inside the container. Follow these steps:

1. Clone the repository and navigate to the finetuned directory containing the Dockerfile and docker-compose.yml.

2. Run the following command to build and start the container:

   docker-compose up --build -d

3. Verify that the docker is running on localhost:11434:

   docker ps

   if result is:

   CONTAINER ID   IMAGE                         COMMAND                  CREATED          STATUS          PORTS                     NAMES
   1e9bff7c2f7b   finetuned-ollama-llm:latest   "/entrypoint.sh"         42 minutes ago   Up 42 minutes   11434/tcp                 compassionate_wing
   

   so run the docker image specifying the port:

   docker run -d -p 11434:11434 finetuned-ollama-llm:latest
   docker ps

   the result will be like:

   CONTAINER ID   IMAGE                         COMMAND                  CREATED         STATUS          PORTS                      NAMES
   df8b31d4ed86   finetuned-ollama-llm:latest   "/entrypoint.sh"         7 seconds ago   Up 7 seconds    0.0.0.0:11434->11434/tcp   eloquent_jennings
   

   check if ollama is runnin in the port 11434:

   curl http://localhost:11434
   

   Result should be:

   Ollama is running
   

This will download the model file, create the model in Ollama, and serve it on port 11434. The necessary directories will be created automatically.

Option 2: Manual Setup

If you prefer to set up the model manually, follow these steps:

1. Download the .gguf file from here.

2. Create a file named Modelfile with the following content:

   FROM ./unsloth.Q4_K_M.gguf

3. Place the downloaded .gguf file in the same folder as the Modelfile.

4. Ensure Ollama is running on your system.

5. Run the following command to create the model in Ollama:

   ollama create llama3.1:8b-brick-v8 -f Modelfile

Once you've set up the model in Ollama, you can use it in your code as follows:

from brickllm.graphs import BrickSchemaGraphLocal

instructions = """
Your job is to generate a RDF graph in Turtle format from a description of energy systems and sensors of a building in the following input, using the Brick ontology.
### Instructions:
- Each subject, object of predicate must start with a @prefix.
- Use the prefix bldg: with IRI <http://my-bldg#> for any created entities.
- Use the prefix brick: with IRI <https://brickschema.org/schema/Brick#> for any Brick entities and relationships used.
- Use the prefix unit: with IRI <http://qudt.org/vocab/unit/> and its ontology for any unit of measure defined.
- When encoding the timeseries ID of the sensor, you must use the following format: ref:hasExternalReference [ a ref:TimeseriesReference ; ref:hasTimeseriesId 'timeseriesID' ].
- When encoding identifiers or external references, such as building/entities IDs, use the following schema: ref:hasExternalReference [ a ref:ExternalReference ; ref:hasExternalReference ‘id/reference’ ].
- When encoding numerical reference, use the schema [brick:value 'value' ; \n brick:hasUnit unit:'unit' ] .
-When encoding coordinates, use the schema brick:coordinates [brick:latitude "lat" ; brick:longitude "long" ].
The response must be the RDF graph that includes all the @prefix of the ontologies used in the triples. The RDF graph must be created in Turtle format. Do not add any other text or comment to the response.
"""

building_description = """
The building (external ref: 'OB103'), with coordinates 33.9614, -118.3531, has a total area of 500 m². It has three zones, each with its own air temperature sensor.
The building has an electrical meter that monitors data of a power sensor. An HVAC equipment serves all three zones and its power usage is measured by a power sensor.

Timeseries IDs and unit of measure of the sensors:
- Building power consumption: '1b3e-29dk-8js7-f54v' in watts.
- HVAC power consumption: '29dh-8ks3-fvjs-d92e' in watts.
- Temperature sensor zone 1: 't29s-jk83-kv82-93fs' in celsius.
- Temperature sensor zone 2: 'f29g-js92-df73-l923' in celsius.
- Temperature sensor zone 3: 'm93d-ljs9-83ks-29dh' in celsius.
"""

# Create an instance of BrickSchemaGraphLocal
brick_graph_local = BrickSchemaGraphLocal(model="llama3.1:8b-brick")

# Display the graph structure
brick_graph_local.display()

# Prepare input data
input_data = {
    "user_prompt": building_description,
    "instructions": instructions
}

# Run the graph
result = brick_graph_local.run(input_data=input_data, stream=False)

# Print the result
print(result)

# Save the result to a file
brick_graph_local.save_ttl_output("my_building_local.ttl")

📖 Documentation

For more detailed information on how to use BrickLLM, please refer to our documentation.

▶️ Web Application

A web app is available to use the library directly through an interface at the following link (). The application can also be used locally as described in the dedicated repository BrickLLM App.

Note: The tool is currently being deployed on our servers and on the MODERATE platform. It will be online shortly !

🤝 Contributing

We welcome contributions to BrickLLM! Please see our contributing guidelines for more information.

📜 License

BrickLLM is released under the BSD-3-Clause License. See the LICENSE file for details.

📧 Contact

For any questions or support, please contact:

• Marco Perini [email protected]
• Daniele Antonucci [email protected]
• Rocco Giudice [email protected]

📝 Citation

Please cite us if you use the library

💙 Acknowledgements

This work was carried out within European projects:

Moderate - Horizon Europe research and innovation programme under grant agreement No 101069834, with the aim of contributing to the development of open products useful for defining plausible scenarios for the decarbonization of the built environment BrickLLM is developed and maintained by the Energy Efficiency in Buildings group at EURAC Research. Thanks to the contribution of:

• Moderate project: Horizon Europe research and innovation programme under grant agreement No 101069834
• Politecnico of Turin, in particular to @Rocco Giudice for his work in developing model generation using local language model

---

Thank you to Brick for the great work it is doing