Flashy
A newer find replaces this one: Flashy. Kept here for reference.
What it is
Flashy is a lightweight Python framework for writing deep learning training loops, designed to be simpler and more hackable than PyTorch-Lightning. It provides a BaseSolver class that handles checkpointing, metric logging, and distributed training utilities, while allowing users to define custom training stages and epochs.
The problem it solves
Existing deep learning frameworks like PyTorch-Lightning are often too complex and rigid for research use cases, making it difficult to implement custom training logic or hack the training loop without fighting the framework's abstractions.
What you could use it for
- Building custom deep learning training scripts with flexible checkpointing and logging
- Implementing distributed training workflows that require complex state management
- Research projects where standard frameworks like PyTorch-Lightning are considered too rigid or over-engineered
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.