Torchboard is a powerful library for real-time neural network modification and monitoring in PyTorch. Torchboard provides an interactive dashboard to tweak model architectures, visualize training metrics, and optimize hyperparameters on the fly.
Install Torch-Board via pip:
pip install torchboard - 📊 Live Metrics Dashboard: Visualize loss, accuracy, gradients, and other metrics in real time through an intuitive interface.
- 🛠️ Interactive Model Editing: Modify layers, activations, hyperparameters, and even stop or adjust training without restarting the training loop.
- 🌐 Automatic Localhost Hosting: The dashboard is automatically hosted on localhost, enabling seamless interaction with your model via a web interface.
- 🔍 Advanced Visualization Tools: Monitor weight distributions, activations, computational graph changes, and other insights like learning rate, accuracy, and loss curves.
- 🤝 Seamless PyTorch Integration: Supports most PyTorch optimizers and integrates effortlessly into existing workflows.
Contributions are welcome! Please submit a pull request or open an issue to add new decorators or suggest improvements.
This project is licensed under the Apache v2.0 License.