Morphik Core

Use the evaluate_result.py script to evaluate a previously created graph against the SciER ground truth. You must provide the --graph-name produced during the creation step.

python evaluate_result.py --graph-name scier_gpt4o_12345678

To manually format all Python files in the project using the project's standard toolchain (isort, black, and ruff), execute the format.sh script located in the scripts/ directory. This ensures consistency with the automated pre-commit hooks.

./scripts/format.sh

To run the @morphik/ui project in development mode:

1. Install dependencies:

npm install

2. Start the development server:

npm run dev

3. Open http://localhost:3000 in your browser.

4. Connect to your Morphik server using a URI obtained from the /local/generate_uri endpoint.

To set up Morphik Core locally, ensure you have Docker and Docker Compose installed, with at least 10GB of free disk space and 8GB+ RAM recommended.

Follow these steps:

1. Clone the repository and enter the directory.
2. Run the build command to initialize all services, including PostgreSQL (with pgvector) and Ollama (with nomic-embed-text and llama3.2 models).
3. Use standard compose commands for lifecycle management.

Note: If you use the ui profile, you must include --profile ui when stopping services.

# Clone and enter directory
git clone https://github.com/morphik-org/morphik-core.git
cd morphik-core

# First-time setup (builds containers and downloads models)
docker compose up --build

# Subsequent runs
docker compose up
docker compose down

# Completely reset (deletes all data and models)
docker compose down -v

To apply a custom configuration file to the morphik service, mount your local file into the container via the docker-compose.yml volumes section.

services:
  morphik:
    volumes:
      - ./my-custom-morphik.toml:/app/morphik.toml

To evaluate a RAG system using the framework, follow these three steps:

1. Run Evaluation: Execute the system implementation to generate raw answers. For the Morphik implementation, run python morphik_eval.py. This produces morphik_results.csv.
2. Judge Results: Use an LLM judge to compare system answers against golden answers. Run python evaluate.py <results_file>.csv. This produces a judged CSV file (e.g., morphik_results_judged.csv).
3. Analyze Results: Perform accuracy and error pattern analysis using python analyze_eval.py <judged_results_file>.csv.

# 1. Run Morphik evaluation
python morphik_eval.py

# 2. Judge Results
python evaluate.py morphik_results.csv

# 3. Analyze Results
python analyze_eval.py morphik_results_judged.csv

Run the scier_evaluation.py script to generate a knowledge graph. Upon completion, the script will output a specific graph name (e.g., scier_gpt4o_12345678) which is required for the evaluation step.

python scier_evaluation.py --model-name gpt4o

To build and start the @morphik/ui application for a production environment, use the following commands:

npm run build
npm start