Back to Search
Start Over
nGPT: Normalized Transformer with Representation Learning on the Hypersphere
- Publication Year :
- 2024
-
Abstract
- We propose a novel neural network architecture, the normalized Transformer (nGPT) with representation learning on the hypersphere. In nGPT, all vectors forming the embeddings, MLP, attention matrices and hidden states are unit norm normalized. The input stream of tokens travels on the surface of a hypersphere, with each layer contributing a displacement towards the target output predictions. These displacements are defined by the MLP and attention blocks, whose vector components also reside on the same hypersphere. Experiments show that nGPT learns much faster, reducing the number of training steps required to achieve the same accuracy by a factor of 4 to 20, depending on the sequence length.
- Subjects :
- Computer Science - Machine Learning
Computer Science - Artificial Intelligence
Subjects
Details
- Database :
- arXiv
- Publication Type :
- Report
- Accession number :
- edsarx.2410.01131
- Document Type :
- Working Paper