Update app.py

app.py

@@ -84,6 +84,10 @@ def generate_answers(img=None, aud = None, q = None, max_tokens = 30):
         print(predicted_caption)
         next_token_embeds = phi2_model.model.embed_tokens(predicted_word_token)
         combined_embeds   = torch.cat([combined_embeds, next_token_embeds], dim=1)
+        del next_token_embeds
+        del predicted_word_token
+        del predicted_word_token_logits
+    del combined_embeds
     predicted_captions_decoded = tokenizer.batch_decode(predicted_caption,ignore_index = 50256)[0]
     predicted_captions_decoded = predicted_captions_decoded.replace("<|endoftext|>","")
     return predicted_captions_decoded