title and summary

llama.py

@@ -9,14 +9,22 @@ from inference import Inference, torch_reseed
 messages = []
 inference = None
 
+# systemmessage at the very begin of the chat. Will be concatenated with the automatic tool usage descriptions
 systemmessage = "Hold a casual conversation with the user. Keep responses short at max 3 sentences."
 
+# system message for role flip so the model automatically answers for the user
 roleflip = {"role": "system", "content": "Keep the conversation going, ask for more information on the subject. Keep messages short at max 1-2 sentences. Do not thank and say goodbye."}
 
+# system messages and user message to bring the model to summarize the entire conversation
+summarize = {"role": "system", "content": "Summarize the conversation as a single, cohesive paragraph. Avoid using any bullet points, numbers, or list formatting. Write in plain text with natural sentences that flow together seamlessly."}
+summarize_user = {"role": "system", "content": "Can you summarize the conversation?"}
+
+# system message to create a conversation title
+title_prompt = {"role": "system", "content": "Please create a very short and descriptive title or label for this conversation. Maximum 2-5 words. Use only plain text, avoid numbering, special characters, or unnecessary formatting-focus on clarity and brevity."}
+
 
 
 register_dummy()
-# tool_functions = [current_time, random_float, random_int]
 
 
 
@@ -126,6 +134,20 @@ def main():
             messages = messages_backup + [last_message]
             append_generate_chat(None)  # 'regular' chatbot answer
 
+        elif input_text.startswith("/summarize"):
+            messages_temp = list(filter(lambda x: x["role"] != "system", messages))
+            messages_temp = [summarize] + messages_temp + [summarize_user]  # copy dict in last instance
+            # messages_temp[-1]["role"] = "user"
+            input_ids = inference.tokenize(messages_temp, tokenize=True, assistant_prefix="The conversation was about ")
+            generated_tokens, full_output = inference.generate_incremental(input_ids)
+
+        elif input_text.startswith("/title"):
+            messages_temp = list(filter(lambda x: x["role"] != "system", messages))
+            messages_temp = [title_prompt] + messages_temp #+ [dict(title)]  # copy dict in last instance
+            messages_temp[-1]["role"] = "user"
+            input_ids = inference.tokenize(messages_temp, tokenize=True, assistant_prefix="Title: ")
+            generated_tokens, full_output = inference.generate_incremental(input_ids)
+
         elif input_text.startswith("/help"):
             print("!<prompt>   answer as 'tool' in <tool_response> tags")
             print("/clear      clear chat history")
@@ -134,6 +156,8 @@ def main():
             print("/more       generate more additional information")
             print("/file       read prompt input from file")
             print("/auto       automatically advance conversation")
+            print("/summarize  generate a summary of the chat")
+            print("/title      generate a title of the chat")
             print("/help       print this message")
             print("")