Merge pull request #2 from eltociear/patch-1

Update README.md

README.md

@@ -49,7 +49,7 @@ Hence, there is no accuracy drop from the memory-efficient reversible backpropag
 
 Here are the [Training/Validation Logs](https://api.wandb.ai/links/action_anticipation/r7k0v3kd) 💯 
 
-👁️ **Note**: The relatively low accuracy is due to difficulty in training vision transformer (reversible or vanilla) from scratch on small datasets like CIFAR-10. Also likely is that6 a much higher accuracy can be achieved with the same code, using a better [chosen model design and optimization paramaters](https://github.com/tysam-code/hlb-CIFAR10). The authors have done no tuning since this repository is meant for understanding code, not pushing performance. 
+👁️ **Note**: The relatively low accuracy is due to difficulty in training vision transformer (reversible or vanilla) from scratch on small datasets like CIFAR-10. Also likely is that6 a much higher accuracy can be achieved with the same code, using a better [chosen model design and optimization parameters](https://github.com/tysam-code/hlb-CIFAR10). The authors have done no tuning since this repository is meant for understanding code, not pushing performance. 
 
 <h2> Running ImageNet, Kinetics-400 and more </h2>