exper
stands as a meticulously crafted Python package, tailored with a focus on expediting the execution of PyTorch-based deep learning experiments, particularly within the realm of graph learning projects. Its lightweight nature doesn't compromise its efficacy; rather, it's designed to enhance the efficiency of your workflow, offering robust support for critical functionalities like Distributed Data Parallel (DDP) Training and Mixed Precision Training. Additionally, it provides a convenient mechanism for logging experiment details and preserving weight files, fostering a seamless and organized experimental process.
-
🔥 PyTorch Integration: At its core,
exper
is built upon PyTorch, harnessing the power of one of the most widely-used deep learning frameworks. This integration ensures compatibility and smooth interoperability with the broader PyTorch ecosystem. -
⚡️ DDP Training Support: Scaling deep learning models across multiple GPUs is a breeze with
exper
. Its robust support for Distributed Data Parallel training ensures that your model training process is not only efficient but also scalable, enabling you to tackle larger datasets and more complex models with ease. -
📝 Experiment Logging: Reproducibility is a cornerstone of scientific research, and
exper
understands its importance. With built-in capabilities for experiment logging, it allows you to effortlessly capture and preserve experiment details, including parameters, configurations, and results. This not only facilitates reproducibility but also simplifies analysis and comparison across different experiments. -
💊 Based on TorchDrug:
exper
draws inspiration from and is derived from the esteemed open-source library, TorchDrug, developed by the Montreal Institute for Learning Algorithms (MILA). Leveraging the solid foundation laid by TorchDrug,exper
inherits reliability and robustness, empowering you to conduct deep learning experiments with confidence.
pip install exper
exper is released under the Apache-2.0 License.
Your contributions are invaluable in making exper even more user-friendly and robust. Feel free to dive in and contribute your code, ideas, or suggestions to the project. Together, let's elevate the experience of deep learning experimentation.