Merge pull request #69 from kookmin-sw/swjeong

커스텀 모델 샘플 작업 및 추론 템플릿 변경

automation/karpenter_node_pool_deploy/IaC/.gitignore

@@ -0,0 +1 @@
+var.tf

automation/karpenter_node_pool_deploy/IaC/var.tf → automation/karpenter_node_pool_deploy/IaC/var.tf.sample

@@ -1,14 +1,14 @@
 variable "region" {
   type    = string
-  default = "ap-northeast-2"
+  default = ""
 }
 
 variable "awscli_profile" {
   type    = string
-  default = "mhsong-swj"
+  default = ""
 }
 
 variable "eks_cluster_name" {
     type = string
-    default = "swj-eks-test"
+    default = ""
 }

example/GAN_Anime/.gitignore

@@ -0,0 +1,3 @@
+input*
+torch.pt
+GAN-anime.zip

example/GAN_Anime/model.py

@@ -0,0 +1,101 @@
+import torch.nn as nn
+
+class ModelClass(nn.Module):
+    def __init__(self):
+        super(ModelClass, self).__init__()
+
+
+        self.convt1=nn.ConvTranspose2d( 100, 512, 4, 1, 0, bias=False)
+        self.btnt1=nn.BatchNorm2d(512)
+        self.relut1=nn.ReLU(True)
+
+            # ``512*4*4``
+        self.convt2=nn.ConvTranspose2d(512,256, 4, 2, 1, bias=False)
+        self.btnt2=nn.BatchNorm2d(256)
+        self.relut2=nn.ReLU(True)
+
+            # ``256*8*8``
+        self.convt3=nn.ConvTranspose2d(256,128, 4, 2, 1, bias=False)
+        self.btnt3=nn.BatchNorm2d(128)
+        self.relut3=nn.ReLU(True)
+
+            # ``128*16*16``
+        self.convt4=nn.ConvTranspose2d(128,64, 4, 2, 1, bias=False)
+        self.btnt4=nn.BatchNorm2d(64)
+        self.relut4=nn.ReLU(True)
+
+            # 64*32*32``
+        self.convt5= nn.ConvTranspose2d( 64, 3, 4, 2, 1, bias=False)
+        self.tan=nn.Tanh()
+
+     # 3*64*64``
+    def forward(self, Input):
+        output=self.convt1(Input)
+        output=self.btnt1(output)
+        output=self.relut1(output)
+
+        output=self.convt2(output)
+        output=self.btnt2(output)
+        output=self.relut2(output)
+
+        output=self.convt3(output)
+        output=self.btnt3(output)
+        output=self.relut3(output)
+
+        output=self.convt4(output)
+        output=self.btnt4(output)
+        output=self.relut4(output)
+
+        output=self.convt5(output)
+        output=self.tan(output)
+
+        return output
+
+class Discriminator(nn.Module):
+    def __init__(self):
+        super(Discriminator, self).__init__()
+
+            # input is `3x 64 x 64``
+        self.conv1=nn.Conv2d(3, 64, 4, 2, 1, bias=False)
+        self.relu1=nn.LeakyReLU(0.2, inplace=True)
+
+            # ``64 x 32 x 32``
+        self.conv2= nn.Conv2d(64, 128, 4, 2, 1, bias=False)
+        self.btn2=nn.BatchNorm2d(128)
+        self.relu2=nn.LeakyReLU(0.2, inplace=True)
+
+            # ``128 x 16 x 16``
+        self.conv3= nn.Conv2d(128,256, 4, 2, 1, bias=False)
+        self.btn3=nn.BatchNorm2d(256)
+        self.relu3=nn.LeakyReLU(0.2, inplace=True)
+
+            # ``256 x 8 x 8``
+        self.conv4= nn.Conv2d(256,512, 4, 2, 1, bias=False)
+        self.btn4=nn.BatchNorm2d(512)
+        self.relu4=nn.LeakyReLU(0.2, inplace=True)
+
+            # 512 x 4 x 4``
+        self.conv5= nn.Conv2d(512, 1, 4, 1, 0, bias=False)
+        self.sig=nn.Sigmoid()
+
+
+    def forward(self, Input):
+        output=self.conv1(Input)
+        output=self.relu1(output)
+
+        output=self.conv2(output)
+        output=self.btn2(output)
+        output=self.relu2(output)
+
+        output=self.conv3(output)
+        output=self.btn3(output)
+        output=self.relu3(output)
+
+        output=self.conv4(output)
+        output=self.btn4(output)
+        output=self.relu4(output)
+
+        output=self.conv5(output)
+        output=self.sig(output)
+
+        return output

inference/template_code/kubernetes_app.py

@@ -37,7 +37,8 @@
 
 try:
     model = ModelClass()
-    model.load_state_dict(torch.load("./model/torch.pt"))
+    if os.path.exists('./model/torch.pt'):
+        model.load_state_dict(torch.load("./model/torch.pt"))
     model.to(device)
 except Exception as e:
     print(f"Model load failed: {e}")

inference/template_code/lambda_app.py

@@ -30,7 +30,8 @@
 
 try:
     model = ModelClass()
-    model.load_state_dict(torch.load("/tmp/model/torch.pt"))
+    if os.path.exists('/tmp/model/torch.pt'):
+        model.load_state_dict(torch.load("/tmp/model/torch.pt", map_location=torch.device('cpu')))
 except Exception as e:
     print(f"Model load failed: {e}")
     os._exit(0)

model_profile/template_code/kubernetes_app.py

@@ -113,11 +113,11 @@ def get_max_used_gpu_memory(device=None):
         value_type = torch.bool
     if model_value_range == "":
         input_data = torch.rand(size=ast.literal_eval(model_input_shape),
-                                dtype=value_type)
+                                dtype=value_type).to(device)
     else:
         value_range = ast.literal_eval(model_value_range)
         input_data = torch.randint(low=value_range[0], high=value_range[1],
-                                   size=ast.literal_eval(model_input_shape))
+                                   size=ast.literal_eval(model_input_shape)).to(device)
 
     torch.cuda.reset_max_memory_allocated(device)
     start_time = time()

model_profile/template_code/requirements_kubernetes.txt

@@ -1,4 +1,5 @@
 torch==2.3.0
 numpy==1.26.4
 requests==2.31.0
-psutil==5.9.8
+psutil==5.9.8
+torchvision==0.15.2