update README

README.md

@@ -4,6 +4,75 @@ This reposity contains the PyTorch training code for the original DeiT models. C
 
 Here, I have build an interface and add some naive methods for add sparsity into the ViT.
 
+## Sparsity NAS Training scripts
+- Normal command
+    - training
+        ```
+        python -m torch.distributed.launch --master_port 29510 --nproc_per_node=2 --use_env main.py \
+        --data-path /dataset/imagenet \
+        --epochs 150 \
+        --pretrained \
+        --lr 5e-5 \ 
+        --min-lr 1e-6 \
+        --nas-mode \
+        --nas-config configs/deit_small_nxm_uniform24.yaml \
+        --nas-test-config 2 4 \
+        --output_dir nas_uniform_24_150epoch \
+        --wandb
+        ```
+    - eval 
+        ```
+        python -m torch.distributed.launch --master_port 29510 --nproc_per_node=2 --use_env main.py \
+        --data-path /dataset/imagenet \
+        --nas-mode \
+        --nas-config configs/deit_small_nxm_uniform24.yaml \
+        --nas-weights nas_uniform_24_150epoch/best_checkpoint.pth \
+        --nas-test-config 2 4 \
+        --eval
+        ```
+- KD command
+    - training
+        ```
+        python -m torch.distributed.launch --master_port 29510 --nproc_per_node=2 --use_env main.py \
+        --data-path /dataset/imagenet \
+        --epochs 150 \
+        --pretrained \\
+        --lr 5e-5 \ 
+        --min-lr 1e-6 \
+        --nas-mode \
+        --nas-config configs/deit_small_nxm_nas_1234.yaml \
+        --nas-test-config 2 4 \
+        --output_dir KD_nas_124+13_150epoch \
+        --teacher-model deit_small_patch16_224 \
+        --distillation-type soft \
+        --distillation-alpha 1.0 \
+        --wandb
+        ```
+    - eval 
+        ```
+        python -m torch.distributed.launch --master_port 29510 --nproc_per_node=2 --use_env main.py \
+        --data-path /dataset/imagenet \
+        --nas-mode \
+        --nas-config configs/deit_small_nxm_uniform24.yaml \
+        --nas-weights KD_nas_124+13_150epoch/checkpoint.pth \
+        --nas-test-config 2 4 \
+        --eval
+        ```
+- Cifar-100 command
+    - training
+        ```
+        python main.py \
+            --model deit_small_patch16_224 \
+            --batch-size 256 \
+            --finetune https://dl.fbaipublicfiles.com/deit/deit_small_patch16_224-cd65a155.pth \
+            --data-set CIFAR \
+            --data-path /dataset/cifar100 \
+            --opt sgd \
+            --weight-decay 1e-4 \
+            --lr 1e-2 \
+            --output_dir deit_s_224_cifar_100 \
+            --epochs 500
+        ```
 
 ## Support Sparsity Searching Algorithm
 Currently, we support the following sparsity strategy:
@@ -23,7 +92,7 @@ We can provide a custom config that define the target sparsity of each layer.
 Currently, we support two kind of sparsity including `nxm` and `unstructuted`.
 User can create a `yaml` file the descibe the detail and pass into the main function by add the `--custom-config [path to config file]` argument when you call the `main.py`
 
-## Example Usage
+## Example Usage (Pruning method)
 To run a DeiT-S with custom configuration and eval the accuracy before finetuning
 ```
 python main.py \