A simple, realtime visualization server for monitoring neural network training performance across frameworks.
Pastalog is a real-time visualization server designed to monitor the training performance of neural networks. It allows developers and researchers to track metrics like loss and accuracy across multiple models as they train, providing immediate visual feedback without interrupting the training process. The tool is framework-agnostic, working with popular libraries like TensorFlow, PyTorch, Keras, and more via a simple API.
Machine learning researchers and developers who need to monitor and compare training metrics across multiple neural network models in real-time, especially those working with frameworks like TensorFlow, PyTorch, or Keras.
Pastalog offers a dead-simple, self-hosted solution for real-time training visualization that works with any deep learning framework. Unlike heavier alternatives, it requires minimal setup, provides an intuitive web interface, and enables direct comparison of models without complex integrations.
Simple, realtime visualization of neural network training performance.
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Installation is straightforward with pip and a few commands, requiring minimal dependencies beyond Node.js, making it quick to deploy for local use.
Works with any deep learning framework via HTTP POST requests or provided APIs, from TensorFlow to Theano, eliminating integration hassles.
Charts update live as metrics are logged, providing immediate feedback on training progress without interrupting scripts, as shown in the demo GIF.
Automatic candlesticking converts dense point clouds into aggregate charts, improving rendering speed and visibility for large datasets, as explained in the usage notes.
Supports panning, zooming, and toggling series visibility directly in the browser, enabling easy model comparison and exploration without extra tools.
Relies on local JSON file storage with a warning to backup data manually, lacking reliability features like replication or cloud sync, which risks data loss in crashes.
Focuses solely on metric plotting, omitting common ML needs such as experiment tracking, hyperparameter logging, or model versioning, limiting its utility for complex workflows.
Requires Node.js 5+, an outdated version unsupported on modern systems, complicating installation and maintenance with potential compatibility issues.
Lacks authentication or encryption, making it unsuitable for shared servers or sensitive data, as anyone can access or modify logs without controls.