header with user input, and grabbed the first response of output_ids[0]

app.py

@@ -34,14 +34,18 @@ def llama_generation(input_text: str,
     Pass input texts, tokenize, output and back to text.
     """
 
-    input_ids = llama_tokenizer.encode(input_text,
+    # Header prompt
+    header = '''Your are a helpful AI called amphisbeana.
+    You will help the user, by giving accurate but creative response'''
+
+    input_ids = llama_tokenizer.encode(input_text + header,
                                        return_tensors='pt').to('cuda')
     
     # llama generation looks for the numeric vectors not the tensors so there is no need for **input_ids rather just input_ids
     output_ids = llama_model.generate(input_ids=input_ids) 
 
     # Decode
-    output_text = llama_tokenizer.decode(output_ids,
+    output_text = llama_tokenizer.decode(output_ids[0],
                                          skip_special_tokens=True)
     
     return output_text