Update app.py

app.py

@@ -75,14 +75,19 @@ def model_generate_ans(img=None,img_audio=None,val_q=None):
             val_q_tokenised = tokenizer(val_q, return_tensors="pt", return_attention_mask=False)['input_ids'].squeeze(0).to(device)
             val_q_embeds    = merged_model.model.embed_tokens(val_q_tokenised).unsqueeze(0)
 
-        val_combined_embeds = torch.empty(5,3,0)
-        
+        val_combined_embeds = []
         if image:
-            val_combined_embeds = torch.cat([val_combined_embeds, val_image_embeds, img_token_embeds], dim=1)
+            #val_combined_embeds = torch.cat([val_combined_embeds, val_image_embeds, img_token_embeds], dim=1)
+            val_combined_embeds.append(val_image_embeds)
+            val_combined_embeds.append(img_token_embeds)
         if img_audio:
-            val_combined_embeds = torch.cat([val_combined_embeds, audio_embeds], dim=1)
+            #val_combined_embeds = torch.cat([val_combined_embeds, audio_embeds], dim=1)
+            val_combined_embeds.append(audio_embeds)
         if val_q:
-            val_combined_embeds = torch.cat([val_combined_embeds, val_q_embeds], dim=1)
+            #val_combined_embeds = torch.cat([val_combined_embeds, val_q_embeds], dim=1)
+            val_combined_embeds.append(val_q_embeds)
+
+        val_combined_embeds = torch.cat(val_combined_embeds,dim=1)
         
         #val_combined_embeds = torch.cat([val_image_embeds, img_token_embeds, val_q_embeds], dim=1) # 4, 69, 2560
         predicted_caption = torch.full((1,max_generate_length),50256).to(device)