Minor fix

app.py

@@ -94,7 +94,7 @@ def infer(message, history):
             predicted_word_token_logits = phi_output_logits[:, -1, :].unsqueeze(1) 
             predicted_word_token = torch.argmax(predicted_word_token_logits, dim = -1) 
             predicted_caption[:,g] = predicted_word_token.view(1,-1)
-            next_token_embeds = model.get_input_embeddings()(prompt_tokens) 
+            next_token_embeds = model.get_input_embeddings()(predicted_word_token) 
             combined_embeds = torch.cat([combined_embeds, next_token_embeds], dim=1)
             
         predicted_captions_decoded = tokenizer.batch_decode(predicted_caption,ignore_index = 50256)[0]