Update app.py

app.py

@@ -147,11 +147,12 @@ def predict1(image_input, question):
         with torch.no_grad():
             # Get image embeddings
             image_embeddings = image_encoder(image)
-            projected_image_embeddings = model.image_projection(image_embeddings)
+            #projected_image_embeddings = model.image_projection(image_embeddings)
 
             # Reshape image embeddings to (batch_size, 1, phi3_embed_dim)
-            projected_image_embeddings = projected_image_embeddings.unsqueeze(1)
-
+            #projected_image_embeddings = projected_image_embeddings.unsqueeze(1)
+            projected_image_embeddings = image_embeddings.unsqueeze(1)
+            
             # Concatenate along the sequence dimension (dim=1)
             extended_attention_mask = torch.cat([torch.ones(projected_image_embeddings.shape[:2], device=encoded["attention_mask"].device), encoded["attention_mask"]], dim=1)
             extended_input_ids = torch.cat([torch.zeros(projected_image_embeddings.shape[:2], dtype=torch.long, device=encoded["input_ids"].device), encoded["input_ids"]], dim=1)