implement ResNet model

train.py

@@ -9,6 +9,7 @@ from label import label_fake, label_real
 from model import mai
 from augmentation import transform_training, transform_validation
 from hyperparams import BATCH_SIZE, EPOCHS
+from resnet import MaiRes
 
 TEST_SIZE = 0.1
 
@@ -37,7 +38,7 @@ def load_data():
         trust_remote_code=True,
         split="train",
     )
-    flickr_dataset = load_dataset("nlphuji/flickr30k", split="test[:50%]")
+    flickr_dataset = load_dataset("nlphuji/flickr30k", split="test[:25%]")
     painting_dataset = load_dataset(
         "keremberke/painting-style-classification", name="full", split="train"
     )
@@ -128,7 +129,7 @@ def train():
     print(f"sample size: {sample_size}")
 
     timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
-    model = mai.cuda()
+    model = MaiRes().cuda()
     loss_fn = torch.nn.BCEWithLogitsLoss()
     optimizer = torch.optim.RMSprop(model.parameters(), lr=0.001)
     best_vloss = 1_000_000.0