File size: 148 Bytes
5fa1a76
 
1
2
The resulting model, the Reformer, performs on par with Transformer models
while being much more memory-efficient and much faster on long sequences.