GPTSwarm

Use pytest to run the test suite. You can run full tests, tests using a mock LLM to avoid API costs, or specific test functions. If using Poetry, prefix commands with poetry run to automatically use the virtual environment.

Common test patterns:

• Mock LLM: Use -m mock_llm to run tests without calling external LLM APIs.
• Logging: Use the -s flag to see logging output in the console.
• Specific Tests: Use -k to filter by function name.

# Quick no-API test (mock LLM) without logging
pytest -m mock_llm test/

# Quick no-API test (mock LLM) with logging
pytest -s -m mock_llm test/

# Standard tests without logging
pytest test/

# Standard tests with logging
pytest -s test/

# Test a specific function by name
pytest -s test/swarm/graph/test_swarm.py -k 'test_raises'

# Running via poetry
poetry run pytest -m mock_llm

To run node optimization that improves the demonstration examples of each node, use the experiments/run_humaneval.py script with the --learn_demonstration flag set to True.

PYTHONPATH=. python experiments/run_humaneval.py --learn_demonstration True

GPTSwarm supports local LLM inference using LM Studio.

1. Download and start the LM Studio desktop app.
2. Load a model from Huggingface.
3. Start the local server.
4. In your GPTSwarm code, set model_name='lmstudio' to route requests to the local server.

To set up the GPTSwarm development environment, clone the repository and use Conda to create a Python 3.10 environment. Then, use Poetry to install the required dependencies.

# Clone the repository
git clone https://github.com/metauto-ai/GPTSwarm.git
cd GPTSwarm/

# Create and activate the environment
conda create -n swarm python=3.10
conda activate swarm

# Install dependencies using poetry
pip install poetry
poetry install

To run general assistant tasks (GAIA), execute the experiments/run_gaia.py script.

PYTHONPATH=. python experiments/run_gaia.py

The GAIA dataset can be obtained from the official Hugging Face repository or via a reconstructed version hosted on Google Drive.

• Official Source: Hugging Face GAIA Benchmark
• Reconstructed Source: Google Drive Link

To run the REINFORCE algorithm for edge optimization using three agents as described in the paper, execute the experiments/run_crosswords.py script.

PYTHONPATH=. python experiments/run_crosswords.py

To build and publish the package using Poetry, you must first configure your PyPI tokens and repository settings. This guide demonstrates publishing to TestPyPI before the main PyPI repository.

# Configure tokens
poetry config pypi-token.pypi "<your-token>"
poetry config repositories.test-pypi https://test.pypi.org/legacy/
poetry config pypi-token.test-pypi "<your-token>"

# Prepare and build
poetry version prerelease
poetry build

# Publish to TestPyPI
poetry publish -r test-pypi

# Publish to main PyPI
poetry publish

# Verify installation from TestPyPI
pip install -i https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple gptswarm==0.1.3a1

You can set up the GPTSwarm development environment using either Poetry (recommended) or Setuptools.

Using Poetry

Use Poetry to install the necessary dependencies and the gptswarm package into your current environment.

Using Setuptools

Install the package in editable mode so that changes to the repository are immediately reflected in the installed package.

### Using Poetry
```bash
# Install needed packages and GPTSwarm
poetry install

# Install needed and uninstall packages not listed in the lock file
poetry install --sync

# Ensure dev packages are installed
poetry install --with=dev --sync

Using Setuptools

# Install as a symbolic link (editable mode)
pip install -e .

# Install with developer tools
pip install -e .[dev]

To measure how much of the codebase is covered by tests, use the coverage tool. This sequence erases previous results, runs the tests, generates an HTML report, and opens it in your browser.

coverage erase
coverage run --source=. -m pytest .
coverage html -i
open htmlcov/index.html

GPTSwarm requires API keys for LLM backends and web search engines. Copy the .env.template file to .env and populate the following keys:

• OPENAI_API_KEY: Required for the OpenAI LLM backend.
• BING_API_KEY: For Bing Web Search.
• GOOGLE_API_KEY: For Google Web Search.
• SEARCHAPI_API_KEY: For SearchAPI Web Search.

Search Engine Priority: The system automatically selects a search engine based on availability in this order:

1. Bing API (if BING_API_KEY is set)
2. Search API (if SEARCHAPI_API_KEY is set)
3. Google API (if GOOGLE_API_KEY is set)

OPENAI_API_KEY=""
BING_API_KEY=""
GOOGLE_API_KEY=""
SEARCHAPI_API_KEY=""