add back files in CycleMLP

image_classification/CycleMLP/README.md

@@ -126,7 +126,6 @@ python main_multi_gpu.py \
 </details>
 
 ## Training
-### Training with single GPU
 To train the CycleMLP model on ImageNet2012 with single GPUs, run the following script using command line:
 ```shell
 sh run_train.sh
@@ -141,9 +140,11 @@ python main_single_gpu.py \
   -data_path='/dataset/imagenet' \
 ```
 
-### Training with multi-GPU
-Run training using multi-GPUs:
+<details>
 
+<summary>
+Run training using multi-GPUs:
+</summary>
 
 
 ```shell
@@ -159,55 +160,9 @@ python main_multi_gpu.py \
     -data_path='/dataset/imagenet' \ 
 ```
 
-
-
-### Training with multi-node
-PaddleVit also supports multi-node distributed training under collective mode.
-
-Suppose you have 2 hosts (denoted as node) with 4 gpus on each machine. 
-Nodes IP addresses are `192.168.0.16` and `192.168.0.17`.
-
-Then some lines of `run_train_multi_node.sh` should be modified:
-```shell
-CUDA_VISIBLE_DEVICES=0,1,2,3 # number of gpus
-
--ips= '192.168.0.16, 192.168.0.17' # seperated by comma
-```
-Run training script in every node:
-```shell
-sh run_train_multi.sh
-```
-
-<details>
-<summary>It is possible to train with multi-node even when you have only one machine</summary>
-
-1. Install docker and paddle. For more details, please refer to
-    [here](https://www.paddlepaddle.org.cn/documentation/docs/zh/install/docker/fromdocker.html).
-
-2. Create a network between docker containers.
-    ```shell
-    docker network create -d bridge paddle_net
-    ```
-3. Create multiple containers as virtual hosts/nodes. Suppose creating 2 containers 
-with 2 gpus on each node.
-    ```shell
-    docker run --name paddle0 -it -d --gpus "device=0,1" --network paddle_net\
-    paddlepaddle/paddle:2.2.0-gpu-cuda10.2-cudnn7 /bin/bash
-    docker run --name paddle1 -it -d --gpus "device=2,3" --network paddle_net\
-    paddlepaddle/paddle:2.2.0-gpu-cuda10.2-cudnn7 /bin/bash
-    ```
-    >   Noted: 
-    >   1. One can assign same gpu device to different containers. But it may occur OOM since multiple models will run on the same gpu. 
-    >   2. One should use `-v` to bind PaddleViT repository to container.
-
-4. Modify `run_train_multi_node.sh` as described above and run the training script on every container.
-
-    >   Noted: One can use `ping` or `ip -a` bash command to check containers' ip addresses. 
-
 </details>
 
 
-
 ## Visualization Attention Map
 **(coming soon)**