Update app.py

app.py

@@ -7,14 +7,13 @@ from PIL import Image
 import os
 import traceback
 import spaces
-import time
 
 # Check if CUDA is available
 device = "cuda" if torch.cuda.is_available() else "cpu"
 print(f"Using device: {device}")
 
 # Load the Byaldi and Qwen2-VL models
-rag_model = RAGMultiModalModel.from_pretrained("vidore/colpali")  # Do not move Byaldi to GPU
+rag_model = RAGMultiModalModel.from_pretrained("vidore/colpali")  # Byaldi model
 qwen_model = Qwen2VLForConditionalGeneration.from_pretrained(
     "Qwen/Qwen2-VL-7B-Instruct", trust_remote_code=True, torch_dtype=torch.bfloat16
 ).to(device)  # Move Qwen2-VL to GPU
@@ -29,15 +28,12 @@ def ocr_and_extract(image):
         temp_image_path = "temp_image.jpg"
         image.save(temp_image_path)
 
-        # Generate a unique index name using the current timestamp
-        unique_index_name = f"image_index_{int(time.time())}"
-
-        # Index the image with Byaldi
+        # Index the image with Byaldi, and force overwrite of the existing index
         rag_model.index(
             input_path=temp_image_path,
-            index_name=unique_index_name,  # Use the unique index name
+            index_name="image_index",  # Reuse the same index
             store_collection_with_index=False,
-            overwrite=True  # Ensure the index is overwritten if it already exists
+            overwrite=True  # Overwrite the index for every new image
         )
 
         # Perform the search query on the indexed image
@@ -79,9 +75,6 @@ def ocr_and_extract(image):
         # Clean up the temporary file
         os.remove(temp_image_path)
 
-        # Clear the index after processing each image to avoid conflicts on future uploads
-        rag_model.clear_index(unique_index_name)
-
         return "\n".join(filtered_output).strip()
 
     except Exception as e: