Update README.md

Author: csuastt

README.md

@@ -246,6 +246,7 @@ Note: If you want to deploy on the Mobile ALOHA robot, don't forget to install t
 - **Use [XFormers](https://github.com/facebookresearch/xformers)**: This library provides optimized transformers with efficient memory usage.
 
 - **Enable Gradient Checkpointing**: Implement `gradient_checkpointing` manually to save memory during backpropagation. See [here](https://deepspeed.readthedocs.io/en/latest/activation-checkpointing.html) for instructions. Once you have successfully implemented this feature, we welcome you to submit a PR👏.
+- **Gradient Accumulation**: Set a larger `--gradient_accumulation_steps=<num_steps>`. This will accumulate the gradients of `<num_steps>` batches for backpropagation. Equivalently, this will increase the batch size by `<num_steps>` times, at the cost of `<num_steps>` times the running time.
 
 ## Citation