Normalization Layer Per-Example Gradients are Sufficient to Predict Gradient Noise Scale in Transformers

Open in new window

The document is on a web server that cannot be reached. Please try again later.