diff --git a/demo/prune/fpgm_mobilenetv1_f-50_train.sh b/demo/prune/fpgm_mobilenetv1_f-50_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..0e84dafd5e9619631927aa118cef02a232016313
--- /dev/null
+++ b/demo/prune/fpgm_mobilenetv1_f-50_train.sh
@@ -0,0 +1,17 @@
+#!/bin/bash  
+export CUDA_VISIBLE_DEVICES=0,1
+export FLAGS_fraction_of_gpu_memory_to_use=0.98
+python train.py \
+    --model="MobileNet" \
+    --pretrained_model="/workspace/models/MobileNetV1_pretrained" \
+    --data="imagenet" \
+    --pruned_ratio=0.3125 \
+    --lr=0.1 \
+    --num_epochs=120 \
+    --test_period=10 \
+    --step_epochs 30 60 90\
+    --l2_decay=3e-5 \
+    --lr_strategy="piecewise_decay" \
+    --criterion="geometry_median" \
+    --model_path="./fpgm_mobilenetv1_models" \
+    2>&1 | tee fpgm_mobilenetv1_train.log
diff --git a/demo/prune/fpgm_mobilenetv2_f-50_train.sh b/demo/prune/fpgm_mobilenetv2_f-50_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..7d0399775f41fb3b762599ab6f2cf63337b7eab4
--- /dev/null
+++ b/demo/prune/fpgm_mobilenetv2_f-50_train.sh
@@ -0,0 +1,17 @@
+#!/bin/bash  
+export CUDA_VISIBLE_DEVICES=0,1
+export FLAGS_fraction_of_gpu_memory_to_use=0.98
+python train.py \
+    --model="MobileNetV2" \
+    --pretrained_model="/workspace/models/MobileNetV2_pretrained" \
+    --data="imagenet" \
+    --pruned_ratio=0.325 \
+    --lr=0.001 \
+    --num_epochs=90 \
+    --test_period=5 \
+    --step_epochs 30 60 80\
+    --l2_decay=1e-4 \
+    --lr_strategy="piecewise_decay" \
+    --criterion="geometry_median" \
+    --model_path="./fpgm_mobilenetv2_models" \
+    2>&1 | tee fpgm_mobilenetv2_train.log
diff --git a/demo/prune/fpgm_resnet34_f-42_train.sh b/demo/prune/fpgm_resnet34_f-42_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..ec24c4d8f13ab2d095e29274326c5136ed240e4d
--- /dev/null
+++ b/demo/prune/fpgm_resnet34_f-42_train.sh
@@ -0,0 +1,12 @@
+#!/bin/bash  
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+export FLAGS_fraction_of_gpu_memory_to_use=0.98
+python train.py \
+    --model="ResNet34" \
+    --pretrained_model="/workspace/models/ResNet34_pretrained" \
+    --data="imagenet" \
+    --pruned_ratio=0.25 \
+    --lr_strategy="cosine_decay" \
+    --criterion="geometry_median" \
+    --model_path="./fpgm_resnet34_025_120_models" \
+    2>&1 | tee fpgm_resnet025_120_train.log
diff --git a/demo/prune/fpgm_resnet34_f-50_train.sh b/demo/prune/fpgm_resnet34_f-50_train.sh
new file mode 100644
index 0000000000000000000000000000000000000000..c181dfb66a1b0fb588cba3868b970f656a73f0d6
--- /dev/null
+++ b/demo/prune/fpgm_resnet34_f-50_train.sh
@@ -0,0 +1,17 @@
+#!/bin/bash  
+export CUDA_VISIBLE_DEVICES=0,1
+export FLAGS_fraction_of_gpu_memory_to_use=0.98
+python train.py \
+    --model="ResNet34" \
+    --pretrained_model="/workspace/models/ResNet34_pretrained" \
+    --data="imagenet" \
+    --pruned_ratio=0.3125 \
+    --lr=0.001 \
+    --num_epochs=70 \
+    --test_period=5 \
+    --step_epochs 30 60 \
+    --l2_decay=1e-4 \
+    --lr_strategy="piecewise_decay" \
+    --criterion="geometry_median" \
+    --model_path="./fpgm_resnet34_models" \
+    2>&1 | tee fpgm_resnet03_train.log