Transformer-Explainability
[CVPR 2021] Official PyTorch implementation for Transformer Interpretability Beyond Attention Visualization, a novel method to visualize classifications by Transformer based networks.
- 2k
- Jupyter Notebook
- MIT
- 3 y ago
[CVPR 2021] Official PyTorch implementation for Transformer Interpretability Beyond Attention Visualization, a novel method to visualize classifications by Transformer based networks.
Attention Saver lets you extract entire attention matrices or row-wise statistics (e.g. entropy) from any HuggingFace causal LLM layer for ultra-long context when using flash-attention without running out of GPU memory.