Slowformer is a transformer framework designed to prioritize:
- Weight Transparency: Gain deep insights into the behavior of weights at every layer, making the training and inference processes more interpretable.
- Monitorability: Seamlessly track weight updates, gradients, and activations with robust logging and visualization tools.
- Deliberate Speed: Crafted for research and experimentation, Slowformer values clarity and explainability over raw computational speed.
- Implement Basic Transformer Model