pamiq-core is a minimal machine learning framework for asynchronous execution of inference and training.
- Simplicity — Clean, intuitive APIs that just work
- Lightweight — Minimal dependencies, maximum performance
- Complete Thread Abstraction — Complex threading handled internally, simple interface externally
When you set out to build a dynamic continuous learning system, PAMIQ Core will be your steadfast foundation.
- 🔄 Parallel Architecture: Simultaneous inference and training in separate threads
- ⚡ Real-time Adaptation: Continuously update models during interaction
- 🧵 Thread-safe Design: Robust synchronization mechanisms for parameter sharing and data transfers
- 🔌 Modular Components: Easy-to-extend agent, environment, and model interfaces
- 🛠️ Comprehensive Tools: Built-in state persistence, time control, and monitoring
- 🏋️ Gymnasium Integration: Seamless compatibility with Gymnasium environments
- 🌍 Cross Platform: Linux is the primary focus, but Windows and macOS are also supported. (However, some older macOS and Windows systems may have significantly less accurate time control.)
- Python 3.12+
- PyTorch (optional, for torch integration)
# Install with pip
pip install pamiq-core
# Optional PyTorch integration
pip install pamiq-core[torch]
# Optional Gymnasium integration
pip install pamiq-core[gym]from pamiq_core import launch, Interaction, LaunchConfig
from your_agent import YourAgent
from your_environment import YourEnvironment
# Create agent-environment interaction
interaction = Interaction(YourAgent(), YourEnvironment())
# Launch the system
launch(
interaction=interaction,
models=your_models,
buffers=your_data_buffers,
trainers=your_trainers,
config=LaunchConfig(
web_api_address=("localhost", 8391),
max_uptime=300.0, # 5 minutes
),
)See the samples directory for complete examples.
Once the system is running, you can connect and control it remotely via the terminal using pamiq-console:
# Connect to local system
pamiq-console --host localhost --port 8391
# Connect to remote system
pamiq-console --host 192.168.1.100 --port 8391
pamiq-core implements a unique architecture that enables autonomous intelligence:
- Concurrent Threads: Separate threads for control, inference, and training
- Parameter Sharing: Thread-safe model parameter synchronization
- Experience Collection: Automatic buffering of data from environments, such as images and audio.
- Continuous Learning: Training models while simultaneously using them for decision making
- State Persistence: Saving and loading system state for resumable operation
Contributions are welcome! Please see CONTRIBUTING.md for guidelines on how to contribute to pamiq-core.
This project is licensed under the MIT License - see the LICENSE file for details.
- pamiq-recorder: Recording library for P-AMI<Q>
- pamiq-io: I/O library for P-AMI<Q>