Fetching the paper…

Random-LTD: Random and Layerwise Token Dropping Brings Efficient Training for Large-scale Transformers · Around