RHAPSODY – Runtime for Heterogeneous APplications, Service Orchestration and DYnamism
A unified runtime for executing AI and HPC workloads on supercomputing infrastructures. RHAPSODY seamlessly integrates traditional scientific computing with AI inference, enabling complex workflows that combine simulation, analysis, and machine learning.
- Unified AI-HPC API: Single interface for compute tasks and AI inference
- Multi-Backend Execution: Run on local machines, HPC clusters (Dragon), or distributed systems (Dask)
- Async-First Design: Native asyncio integration for efficient task orchestration
- Integratable Design: RHAPSODY is designed to be integratable with existing workflows and tools such as AsyncFlow and LangGraph/FlowGentic.
- Scale-Ready: Scale your workload and workflows to thousands of tasks and nodes.
import asyncio
from rhapsody.api import Session, ComputeTask, AITask
from rhapsody.backends import DragonExecutionBackend, DragonVllmInferenceBackend
from rhapsody.backends.ai.config import ModelConfig
async def main():
# Initialize backends
hpc_backend = await DragonExecutionBackend(name="hpc")
ai_backend = await DragonVllmInferenceBackend(
name="vllm", model=ModelConfig(model_name="Qwen2.5-7B", hf_token="...", tp_size=1)
)
# Create session with multiple backends
async with Session(backends=[hpc_backend, ai_backend]) as session:
# HPC simulation task
simulation = ComputeTask(
executable="./simulate",
arguments=["--config", "params.yaml"],
backend=hpc_backend.name
)
# AI analysis task
analysis = AITask(
prompt="Analyze the simulation results and identify key patterns...",
backend=ai_backend.name
)
# Submit and execute
await session.submit_tasks([simulation, analysis])
# Wait for completion (tasks are awaitable!)
sim_result = await simulation
ai_result = await analysis
print(f"Simulation: {sim_result['stdout']}")
print(f"AI Analysis: {ai_result['response']}")
asyncio.run(main())# Basic installation
pip install rhapsody-py
# With specific backends
pip install rhapsody-py[dask] # Dask distributed computing
pip install rhapsody-py[dragon] # Dragon runtime (Python 3.10-3.12)
# Development
pip install rhapsody-py[dev]- Full Documentation: https://radical-cybertools.github.io/rhapsody/
- API Reference: https://radical-cybertools.github.io/rhapsody/api/
- Examples: See
examples/directory
We welcome contributions! Please see our Contributing Guide for details.
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Make your changes
- Add tests for new functionality
- Ensure all tests pass (
make test-regular) - Run code quality checks (
pre-commit run --all-files) - Commit your changes (
git commit -m 'Add amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Open a Pull Request
Please use the GitHub issue tracker to report bugs or request features.
RHAPSODY is licensed under the MIT License.
RHAPSODY is a collaborative project between two teams:
- RADICAL Research Group — Rutgers University, New Jersey: core architecture, API design, session management, telemetry, and backend abstraction.
- Hewlett Packard Enterprise (HPE) — USA / Canada: DragonHPC runtime integration and Dragon backend development.
See the full Team page for details.
- AsyncFlow: Asynchronous workflow management
RHAPSODY is supported by the National Science Foundation (NSF) under Award ID 2103986. This collaborative project aims to advance the state-of-the-art in heterogeneous workflow execution for scientific computing.
If you use RHAPSODY in your research, please cite:
@software{rhapsody2024,
title={RHAPSODY: Runtime for Heterogeneous Applications, Service Orchestration and Dynamism},
author={RADICAL Research Group, Rutgers University and Hewlett Packard Enterprise},
year={2024},
url={https://github.com/radical-cybertools/rhapsody},
version={0.1.0}
}