Update app.py

app.py

@@ -4,6 +4,7 @@ from transformers import CLIPTokenizerFast, CLIPProcessor, CLIPModel
 import torch
 from tqdm.auto import tqdm
 import numpy as np
+import time
 
 device = 'cpu' # 'cuda' if torch.cuda.is_available() else "cpu"
 model_id = 'openai/clip-vit-base-patch32'
@@ -30,31 +31,22 @@ def embedding_input(text_input):
 
 def embedding_img():
     global images
-    sample_idx= np.random.randint(0, len(imagenette)+1, 100).tolist()
-    images = [imagenette[i]['image'] for i in sample_idx]
-    batch_sie = 5
-    image_arr = None
-    for i in tqdm(range(0, len(images), batch_sie)):
-        batch = images[i:i+batch_sie]
+    img_batch = imagenette['image']
 
-        batch = processor(
-            text = None,
-            images = batch,
-            return_tensors= 'pt',
-            padding = True
-        )['pixel_values'].to(device)
-        batch_emb = model.get_image_features(pixel_values = batch)
-        batch_emb = batch_emb.squeeze(0)
-        batch_emb = batch_emb.cpu().detach().numpy()
-        if image_arr is None:
-            image_arr = batch_emb
-            
-        else:
-            image_arr = np.concatenate((image_arr, batch_emb), axis = 0)
-        return image_arr
+    images = processor(
+        text = None,
+        images = img_batch,
+        return_tensors = 'pt'
+    )['pixel_values'].to(device)
+    batch_emb = model.get_image_features(pixel_values =img_batch)
+    batch_emb = batch_emb.squeeze(0)
+    image_arr = batch_emb.cpu().detach().numpy()
+
+    return image_arr
 
 def norm_val(text_input):
     image_arr = embedding_img()
+    time.sleep(5)
     text_emb = embedding_input(text_input)
 
     image_arr = (image_arr.T / np.linalg.norm(image_arr, axis = 1)).T