Update src/txagent/txagent.py

src/txagent/txagent.py

@@ -67,42 +67,57 @@ class TxAgent:
         self.enable_checker = enable_checker
         self.additional_default_tools = additional_default_tools
         self.print_self_values()
+        logger.info("TxAgent initialized with model_name=%s, rag_model_name=%s", model_name, rag_model_name)
 
     def init_model(self):
-        self.load_models()
-        self.load_tooluniverse()
-        self.load_tool_desc_embedding()
+        logger.info("Initializing model: %s", self.model_name)
+        try:
+            self.load_models()
+            self.load_tooluniverse()
+            self.load_tool_desc_embedding()
+            logger.info("Model initialization complete")
+        except Exception as e:
+            logger.error("Failed to initialize model: %s", e, exc_info=True)
+            raise
 
     def print_self_values(self):
         for attr, value in self.__dict__.items():
-            print(f"{attr}: {value}")
+            logger.debug("%s: %s", attr, value)
 
     def load_models(self, model_name=None):
         if model_name is not None:
             if model_name == self.model_name:
+                logger.debug("Model %s already loaded", model_name)
                 return f"The model {model_name} is already loaded."
             self.model_name = model_name
 
+        logger.debug("Loading model %s", self.model_name)
         self.model = LLM(model=self.model_name)
         self.chat_template = Template(self.model.get_tokenizer().chat_template)
         self.tokenizer = self.model.get_tokenizer()
-
+        logger.info("Model %s loaded successfully", self.model_name)
         return f"Model {model_name} loaded successfully."
 
     def load_tooluniverse(self):
+        logger.debug("Loading tool universe")
         self.tooluniverse = ToolUniverse(tool_files=self.tool_files_dict)
         self.tooluniverse.load_tools()
         special_tools = self.tooluniverse.prepare_tool_prompts(
             self.tooluniverse.tool_category_dicts["special_tools"])
         self.special_tools_name = [tool['name'] for tool in special_tools]
+        logger.debug("Tool universe loaded with %d special tools", len(self.special_tools_name))
 
     def load_tool_desc_embedding(self):
+        logger.debug("Loading tool description embeddings")
         self.rag_model.load_tool_desc_embedding(self.tooluniverse)
+        logger.debug("Tool description embeddings loaded")
 
     def rag_infer(self, query, top_k=5):
+        logger.debug("Running RAG inference with query: %s", query[:50])
         return self.rag_model.rag_infer(query, top_k)
 
     def initialize_tools_prompt(self, call_agent, call_agent_level, message):
+        logger.debug("Initializing tools prompt, call_agent=%s, level=%d", call_agent, call_agent_level)
         picked_tools_prompt = []
         picked_tools_prompt = self.add_special_tools(
             picked_tools_prompt, call_agent=call_agent)
@@ -114,9 +129,11 @@ class TxAgent:
         if not call_agent:
             picked_tools_prompt += self.tool_RAG(
                 message=message, rag_num=self.init_rag_num)
+        logger.debug("Tools prompt initialized with %d tools", len(picked_tools_prompt))
         return picked_tools_prompt, call_agent_level
 
     def initialize_conversation(self, message, conversation=None, history=None):
+        logger.debug("Initializing conversation with message: %s", message[:50])
         if conversation is None:
             conversation = []
 
@@ -125,7 +142,7 @@ class TxAgent:
         if history is not None:
             if len(history) == 0:
                 conversation = []
-                print("clear conversation successfully")
+                logger.debug("Cleared conversation")
             else:
                 for i in range(len(history)):
                     if history[i]['role'] == 'user':
@@ -139,7 +156,7 @@ class TxAgent:
                             {"role": "assistant", "content": history[i]['content']})
 
         conversation.append({"role": "user", "content": message})
-
+        logger.debug("Conversation initialized with %d messages", len(conversation))
         return conversation
 
     def tool_RAG(self, message=None,
@@ -147,6 +164,7 @@ class TxAgent:
                  existing_tools_prompt=[],
                  rag_num=5,
                  return_call_result=False):
+        logger.debug("Running tool RAG, message=%s, rag_num=%d", message[:50] if message else None, rag_num)
         extra_factor = 30
         if picked_tool_names is None:
             assert picked_tool_names is not None or message is not None
@@ -165,39 +183,43 @@ class TxAgent:
             picked_tools)
         if return_call_result:
             return picked_tools_prompt, picked_tool_names
+        logger.debug("Tool RAG returned %d tools", len(picked_tools_prompt))
         return picked_tools_prompt
 
     def add_special_tools(self, tools, call_agent=False):
+        logger.debug("Adding special tools, call_agent=%s", call_agent)
         if self.enable_finish:
             tools.append(self.tooluniverse.get_one_tool_by_one_name(
                 'Finish', return_prompt=True))
-            print("Finish tool is added")
+            logger.debug("Finish tool added")
         if call_agent:
             tools.append(self.tooluniverse.get_one_tool_by_one_name(
                 'CallAgent', return_prompt=True))
-            print("CallAgent tool is added")
+            logger.debug("CallAgent tool added")
         else:
             if self.enable_rag:
                 tools.append(self.tooluniverse.get_one_tool_by_one_name(
                     'Tool_RAG', return_prompt=True))
-                print("Tool_RAG tool is added")
+                logger.debug("Tool_RAG tool added")
 
             if self.additional_default_tools is not None:
                 for each_tool_name in self.additional_default_tools:
                     tool_prompt = self.tooluniverse.get_one_tool_by_one_name(
                         each_tool_name, return_prompt=True)
                     if tool_prompt is not None:
-                        print(f"{each_tool_name} tool is added")
+                        logger.debug("%s tool added", each_tool_name)
                         tools.append(tool_prompt)
         return tools
 
     def add_finish_tools(self, tools):
+        logger.debug("Adding finish tools")
         tools.append(self.tooluniverse.get_one_tool_by_one_name(
             'Finish', return_prompt=True))
-        print("Finish tool is added")
+        logger.debug("Finish tool added")
         return tools
 
     def set_system_prompt(self, conversation, sys_prompt):
+        logger.debug("Setting system prompt")
         if len(conversation) == 0:
             conversation.append(
                 {"role": "system", "content": sys_prompt})
@@ -213,6 +235,7 @@ class TxAgent:
                           call_agent_level=None,
                           temperature=None):
 
+        logger.debug("Running function call with input: %s", fcall_str[:50])
         function_call_json, message = self.tooluniverse.extract_function_call_json(
             fcall_str, return_message=return_message, verbose=False)
         call_results = []
@@ -220,7 +243,7 @@ class TxAgent:
         if function_call_json is not None:
             if isinstance(function_call_json, list):
                 for i in range(len(function_call_json)):
-                    print("\033[94mTool Call:\033[0m", function_call_json[i])
+                    logger.debug("Tool Call: %s", function_call_json[i])
                     if function_call_json[i]["name"] == 'Finish':
                         special_tool_call = 'Finish'
                         break
@@ -255,7 +278,7 @@ class TxAgent:
 
                     call_id = self.tooluniverse.call_id_gen()
                     function_call_json[i]["call_id"] = call_id
-                    print("\033[94mTool Call Result:\033[0m", call_result)
+                    logger.debug("Tool Call Result: %s", call_result)
                     call_results.append({
                         "role": "tool",
                         "content": json.dumps({"tool_name": function_call_json[i]["name"], "content": call_result, "call_id": call_id})
@@ -271,7 +294,7 @@ class TxAgent:
             "content": message.strip(),
             "tool_calls": json.dumps(function_call_json)
         }] + call_results
-
+        logger.debug("Function call completed, returning %d messages", len(revised_messages))
         return revised_messages, existing_tools_prompt, special_tool_call
 
     def run_function_call_stream(self, fcall_str,
@@ -283,7 +306,7 @@ class TxAgent:
                                  temperature=None,
                                  return_gradio_history=True):
 
-        logger.debug(f"Running function call stream with input: {fcall_str[:100]}...")
+        logger.debug("Running function call stream with input: %s", fcall_str[:50])
         function_call_json, message = self.tooluniverse.extract_function_call_json(
             fcall_str, return_message=return_message, verbose=False)
         call_results = []
@@ -303,7 +326,7 @@ class TxAgent:
 
         if isinstance(function_call_json, list):
             for i in range(len(function_call_json)):
-                logger.debug(f"Processing tool call: {function_call_json[i]}")
+                logger.debug("Processing tool call: %s", function_call_json[i])
                 if function_call_json[i]["name"] == 'Finish':
                     special_tool_call = 'Finish'
                     break
@@ -361,12 +384,14 @@ class TxAgent:
         }] + call_results
 
         if return_gradio_history:
-            logger.debug(f"Yielding gradio history with {len(gradio_history)} entries")
+            logger.debug("Yielding gradio history with %d entries", len(gradio_history))
             yield revised_messages, existing_tools_prompt or [], special_tool_call, gradio_history
         else:
             yield revised_messages, existing_tools_prompt or [], special_tool_call
+        logger.debug("Function call stream completed")
 
     def get_answer_based_on_unfinished_reasoning(self, conversation, temperature, max_new_tokens, max_token, outputs=None):
+        logger.debug("Forcing answer due to unfinished reasoning")
         if conversation[-1]['role'] == 'assistant':
             conversation.append(
                 {'role': 'tool', 'content': 'Errors occurred; provide a detailed final answer based on current information.'})
@@ -378,7 +403,7 @@ class TxAgent:
                                           output_begin_string='[FinalAnswer]',
                                           skip_special_tokens=True,
                                           max_new_tokens=max_new_tokens, max_token=max_token)
-        logger.debug(f"Forced finish output: {last_outputs_str[:100]}...")
+        logger.debug("Forced finish output: %s", last_outputs_str[:100])
         return last_outputs_str
 
     def run_multistep_agent(self, message: str,
@@ -388,7 +413,7 @@ class TxAgent:
                             max_round: int = 20,
                             call_agent=False,
                             call_agent_level=0) -> str:
-        print("\033[1;32;40mstart\033[0m")
+        logger.info("Starting multistep agent with message: %s", message[:50])
         picked_tools_prompt, call_agent_level = self.initialize_tools_prompt(
             call_agent, call_agent_level, message)
         conversation = self.initialize_conversation(message)
@@ -407,6 +432,7 @@ class TxAgent:
         try:
             while next_round and current_round < max_round:
                 current_round += 1
+                logger.debug("Round %d", current_round)
                 if len(outputs) > 0:
                     function_call_messages, picked_tools_prompt, special_tool_call = self.run_function_call(
                         last_outputs, return_message=True,
@@ -424,12 +450,12 @@ class TxAgent:
                                 function_call_messages[0]['content'])
                         content = function_call_messages[0]['content']
                         if content is None:
+                            logger.warning("No content after Finish tool call")
                             return "❌ No content returned after Finish tool call."
+                        logger.debug("Returning final content: %s", content[:50])
                         return content.split('[FinalAnswer]')[-1]
 
                     if (self.enable_summary or token_overflow) and not call_agent:
-                        if token_overflow:
-                            print("token_overflow, using summary")
                         enable_summary = True
                     last_status = self.function_result_summary(
                         conversation, status=last_status, enable_summary=enable_summary)
@@ -440,14 +466,14 @@ class TxAgent:
                             function_call_messages))
                     else:
                         next_round = False
-                        conversation.extend(
-                            [{"role": "assistant", "content": ''.join(last_outputs)}])
-                        return ''.join(last_outputs).replace("</s>", "")
+                        content = ''.join(last_outputs).replace("</s>", "")
+                        logger.debug("Returning content: %s", content[:50])
+                        return content
+
                 if self.enable_checker:
                     good_status, wrong_info = checker.check_conversation()
                     if not good_status:
-                        print(
-                            "Internal error in reasoning: " + wrong_info)
+                        logger.warning("Internal error in reasoning: %s", wrong_info)
                         break
                 last_outputs = []
                 outputs.append("### TxAgent:\n")
@@ -458,7 +484,7 @@ class TxAgent:
                                                                   max_new_tokens=max_new_tokens, max_token=max_token,
                                                                   check_token_status=True)
                 if last_outputs_str is None:
-                    print("The number of tokens exceeds the maximum limit.")
+                    logger.warning("Token limit exceeded")
                     if self.force_finish:
                         return self.get_answer_based_on_unfinished_reasoning(conversation, temperature, max_new_tokens, max_token)
                     else:
@@ -466,20 +492,22 @@ class TxAgent:
                 else:
                     last_outputs.append(last_outputs_str)
             if max_round == current_round:
-                print("The number of rounds exceeds the maximum limit!")
+                logger.warning("Max rounds exceeded")
             if self.force_finish:
                 return self.get_answer_based_on_unfinished_reasoning(conversation, temperature, max_new_tokens, max_token)
             else:
+                logger.debug("No output due to max rounds")
                 return None
 
         except Exception as e:
-            print(f"Error: {e}")
+            logger.error("Error in multistep agent: %s", e, exc_info=True)
             if self.force_finish:
                 return self.get_answer_based_on_unfinished_reasoning(conversation, temperature, max_new_tokens, max_token)
             else:
                 return None
 
     def build_logits_processor(self, messages, llm):
+        logger.debug("Building logits processor")
         tokenizer = llm.get_tokenizer()
         if self.avoid_repeat and len(messages) > 2:
             assistant_messages = []
@@ -491,14 +519,14 @@ class TxAgent:
             forbidden_ids = [tokenizer.encode(
                 msg, add_special_tokens=False) for msg in assistant_messages]
             return [NoRepeatSentenceProcessor(forbidden_ids, 5)]
-        else:
-            return None
+        return None
 
     def llm_infer(self, messages, temperature=0.1, tools=None,
                   output_begin_string=None, max_new_tokens=2048,
                   max_token=None, skip_special_tokens=True,
                   model=None, tokenizer=None, terminators=None, seed=None, check_token_status=False):
 
+        logger.debug("Running LLM inference with %d messages", len(messages))
         if model is None:
             model = self.model
 
@@ -519,16 +547,15 @@ class TxAgent:
             input_tokens = self.tokenizer.encode(prompt, return_tensors="pt")[0]
             num_input_tokens = len(input_tokens)
             if num_input_tokens > max_token:
-                logger.info(f"Number of input tokens before inference: {num_input_tokens}")
-                logger.info("The number of tokens exceeds the maximum limit!!!!")
+                logger.info("Input tokens: %d, max_token: %d", num_input_tokens, max_token)
                 max_prompt_tokens = max_token - max_new_tokens - 100
                 if max_prompt_tokens > 0:
                     truncated_input = self.tokenizer.decode(input_tokens[:max_prompt_tokens])
                     prompt = truncated_input
-                    logger.info(f"Prompt truncated to {len(self.tokenizer.encode(prompt, return_tensors='pt')[0])} tokens")
+                    logger.info("Truncated to %d tokens", len(self.tokenizer.encode(prompt, return_tensors='pt')[0]))
                     token_overflow = True
                 else:
-                    logger.warning("Max prompt tokens too small, cannot truncate effectively")
+                    logger.warning("Cannot truncate effectively")
                     torch.cuda.empty_cache()
                     gc.collect()
                     return None, token_overflow
@@ -538,7 +565,7 @@ class TxAgent:
             sampling_params=sampling_params,
         )
         output = output[0].outputs[0].text
-        print("\033[92m" + output + "\033[0m")
+        logger.debug("LLM output: %s", output[:50])
         if check_token_status and max_token is not None:
             return output, token_overflow
 
@@ -549,7 +576,7 @@ class TxAgent:
                        max_new_tokens: int,
                        max_token: int) -> str:
 
-        print("\033[1;32;40mstart self agent\033[0m")
+        logger.info("Starting self agent with message: %s", message[:50])
         conversation = []
         conversation = self.set_system_prompt(conversation, self.self_prompt)
         conversation.append({"role": "user", "content": message})
@@ -563,7 +590,7 @@ class TxAgent:
                        max_new_tokens: int,
                        max_token: int) -> str:
 
-        print("\033[1;32;40mstart chat agent\033[0m")
+        logger.info("Starting chat agent with message: %s", message[:50])
         conversation = []
         conversation = self.set_system_prompt(conversation, self.chat_prompt)
         conversation.append({"role": "user", "content": message})
@@ -578,7 +605,7 @@ class TxAgent:
                          max_new_tokens: int,
                          max_token: int) -> str:
 
-        print("\033[1;32;40mstart format agent\033[0m")
+        logger.info("Starting format agent")
         if '[FinalAnswer]' in answer:
             possible_final_answer = answer.split("[FinalAnswer]")[-1]
         elif "\n\n" in answer:
@@ -588,12 +615,13 @@ class TxAgent:
         if len(possible_final_answer) == 1:
             choice = possible_final_answer[0]
             if choice in ['A', 'B', 'C', 'D', 'E']:
+                logger.debug("Returning choice: %s", choice)
                 return choice
         elif len(possible_final_answer) > 1:
             if possible_final_answer[1] == ':':
                 choice = possible_final_answer[0]
                 if choice in ['A', 'B', 'C', 'D', 'E']:
-                    print("choice", choice)
+                    logger.debug("Returning choice: %s", choice)
                     return choice
 
         conversation = []
@@ -611,7 +639,7 @@ class TxAgent:
                           temperature: float,
                           max_new_tokens: int,
                           max_token: int) -> str:
-        print("\033[1;32;40mSummarized Tool Result:\033[0m")
+        logger.info("Running summary agent")
         generate_tool_result_summary_training_prompt = """Thought and function calls: 
 {thought_calls}
 Function calls' responses:
@@ -632,9 +660,11 @@ Generate **one summarized sentence** about "function calls' responses" with nece
 
         if '[' in output:
             output = output.split('[')[0]
+        logger.debug("Summary output: %s", output)
         return output
 
     def function_result_summary(self, input_list, status, enable_summary):
+        logger.debug("Running function result summary, enable_summary=%s", enable_summary)
         if 'tool_call_step' not in status:
             status['tool_call_step'] = 0
 
@@ -682,13 +712,14 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                         this_thought_calls = None
                     else:
                         if len(function_response) != 0:
-                            print("internal summary")
+                            logger.debug("Generating internal summary")
                             status['summarized_step'] += 1
                             result_summary = self.run_summary_agent(
                                 thought_calls=this_thought_calls,
                                 function_response=function_response,
                                 temperature=0.1,
                                 max_new_tokens=1024,
+                                run_gradio_chat
                                 max_token=99999
                             )
 
@@ -729,15 +760,18 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                 last_call_idx+1, {'role': 'tool', 'content': result_summary})
             status['summarized_index'] = last_call_idx + 2
 
+        logger.debug("Function result summary completed")
         return status
 
     def update_parameters(self, **kwargs):
+        logger.debug("Updating parameters: %s", kwargs)
         for key, value in kwargs.items():
             if hasattr(self, key):
                 setattr(self, key, value)
 
         updated_attributes = {key: value for key,
                               value in kwargs.items() if hasattr(self, key)}
+        logger.debug("Updated attributes: %s", updated_attributes)
         return updated_attributes
 
     def run_gradio_chat(self, message: str,
@@ -763,45 +797,49 @@ Generate **one summarized sentence** about "function calls' responses" with nece
         Returns:
             str: Final assistant message.
         """
-        logger.debug(f"[TxAgent] Chat started, message: {message[:100]}...")
-        print("\033[1;32;40m[TxAgent] Chat started\033[0m")
-
-        if not message or len(message.strip()) < 5:
-            yield "Please provide a valid message or upload files to analyze."
-            return "Invalid input."
-
-        if message.startswith("[\U0001f9f0 Tool_RAG") or message.startswith("⚒️"):
-            return ""
-
-        outputs = []
-        last_outputs = []
+        logger.info("[TxAgent] Chat started with message: %s", message[:100])
+        logger.debug("Initial history: %s", [msg.content[:50] for msg in history] if history else [])
 
-        picked_tools_prompt, call_agent_level = self.initialize_tools_prompt(
-            call_agent,
-            call_agent_level,
-            message)
+        # Yield initial message to ensure UI updates
+        history.append(ChatMessage(role="assistant", content="Starting analysis..."))
+        yield history
+        logger.debug("Yielded initial history")
 
-        conversation = self.initialize_conversation(
-            message,
-            conversation=conversation,
-            history=history)
-        history = []  # Reset history to avoid duplication
+        try:
+            if not message or len(message.strip()) < 5:
+                logger.warning("Invalid message detected")
+                history.append(ChatMessage(role="assistant", content="Please provide a valid message or upload files to analyze."))
+                yield history
+                return "Invalid input."
 
-        next_round = True
-        function_call_messages = []
-        current_round = 0
-        enable_summary = False
-        last_status = {}
-        token_overflow = False
+            if message.startswith("[\U0001f9f0 Tool_RAG") or message.startswith("⚒️"):
+                logger.debug("Skipping tool-related message")
+                yield history
+                return ""
+
+            outputs = []
+            last_outputs = []
+            picked_tools_prompt, call_agent_level = self.initialize_tools_prompt(
+                call_agent, call_agent_level, message)
+            conversation = self.initialize_conversation(
+                message, conversation=conversation, history=history)
+            history = []  # Reset history to avoid duplication
+            logger.debug("Conversation initialized with %d messages", len(conversation))
+
+            next_round = True
+            function_call_messages = []
+            current_round = 0
+            enable_summary = False
+            last_status = {}
+            token_overflow = False
 
-        if self.enable_checker:
-            checker = ReasoningTraceChecker(
-                message, conversation, init_index=len(conversation))
+            if self.enable_checker:
+                checker = ReasoningTraceChecker(
+                    message, conversation, init_index=len(conversation))
 
-        try:
             while next_round and current_round < max_round:
                 current_round += 1
-                logger.debug(f"Round {current_round}, conversation length: {len(conversation)}")
+                logger.debug("Round %d, conversation length: %d", current_round, len(conversation))
 
                 if last_outputs:
                     function_call_result = yield from self.run_function_call_stream(
@@ -812,13 +850,11 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                         call_agent_level=call_agent_level,
                         temperature=temperature)
 
-                    # Ensure function_call_result is valid
                     if not function_call_result:
                         logger.warning("Empty result from run_function_call_stream")
-                        error_msg = "Error: Tool call processing failed."
-                        history.append(ChatMessage(role="assistant", content=error_msg))
+                        history.append(ChatMessage(role="assistant", content="Error: Tool call processing failed."))
                         yield history
-                        return error_msg
+                        return "Error: Tool call processing failed."
 
                     function_call_messages, picked_tools_prompt, special_tool_call, current_gradio_history = function_call_result
 
@@ -830,9 +866,11 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                             unique_history.append(msg)
                             seen_contents.add(msg.content)
                     history.extend(unique_history)
+                    logger.debug("Extended history with %d unique messages", len(unique_history))
 
                     if special_tool_call == 'Finish' and function_call_messages:
                         history.append(ChatMessage(role="assistant", content=function_call_messages[0]['content']))
+                        logger.debug("Yielding final history after Finish: %s", function_call_messages[0]['content'][:50])
                         yield history
                         next_round = False
                         conversation.extend(function_call_messages)
@@ -841,6 +879,7 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                     elif special_tool_call in ['RequireClarification', 'DirectResponse']:
                         last_msg = history[-1] if history else ChatMessage(role="assistant", content="Response needed.")
                         history.append(ChatMessage(role="assistant", content=last_msg.content))
+                        logger.debug("Yielding history for special tool: %s", last_msg.content[:50])
                         yield history
                         next_round = False
                         return last_msg.content
@@ -849,8 +888,7 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                         enable_summary = True
 
                     last_status = self.function_result_summary(
-                        conversation, status=last_status,
-                        enable_summary=enable_summary)
+                        conversation, status=last_status, enable_summary=enable_summary)
 
                     if function_call_messages:
                         conversation.extend(function_call_messages)
@@ -859,13 +897,14 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                         content = ''.join(last_outputs).replace("</s>", "")
                         history.append(ChatMessage(role="assistant", content=content))
                         conversation.append({"role": "assistant", "content": content})
+                        logger.debug("Yielding history with content: %s", content[:50])
                         yield history
                         return content
 
                 if self.enable_checker:
                     good_status, wrong_info = checker.check_conversation()
                     if not good_status:
-                        logger.warning(f"Checker flagged reasoning error: {wrong_info}")
+                        logger.warning("Checker flagged error: %s", wrong_info)
                         break
 
                 last_outputs = []
@@ -879,10 +918,11 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                     seed=seed,
                     check_token_status=True)
 
-                logger.debug(f"llm_infer output: {last_outputs_str[:100] if last_outputs_str else None}, token_overflow: {token_overflow}")
+                logger.debug("llm_infer output: %s, token_overflow: %s", 
+                            last_outputs_str[:50] if last_outputs_str else None, token_overflow)
 
                 if last_outputs_str is None:
-                    logger.warning("llm_infer returned None, likely due to token overflow")
+                    logger.warning("llm_infer returned None")
                     error_msg = "Error: Unable to generate response due to token limit. Please reduce input size."
                     history.append(ChatMessage(role="assistant", content=error_msg))
                     yield history
@@ -899,16 +939,18 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                     final_thought, final_answer = parts if len(parts) == 2 else (last_thought, "")
                     history.append(ChatMessage(role="assistant", content=final_thought.strip()))
                     history.append(ChatMessage(role="assistant", content="**🧠 Final Analysis:**\n" + final_answer.strip()))
+                    logger.debug("Yielding final analysis: %s", final_answer[:50])
                     yield history
                     next_round = False
                 else:
                     history.append(ChatMessage(role="assistant", content=last_thought))
+                    logger.debug("Yielding intermediate history: %s", last_thought[:50])
                     yield history
 
                 last_outputs.append(last_outputs_str)
 
             if next_round:
-                logger.info("Max rounds reached, forcing finish")
+                logger.info("Max rounds reached")
                 if self.force_finish:
                     last_outputs_str = self.get_answer_based_on_unfinished_reasoning(
                         conversation, temperature, max_new_tokens, max_token)
@@ -919,17 +961,20 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                         history.append(ChatMessage(role="assistant", content="**🧠 Final Analysis:**\n" + final_answer.strip()))
                     else:
                         history.append(ChatMessage(role="assistant", content=last_outputs_str.strip()))
+                    logger.debug("Yielding forced final history")
                     yield history
                 else:
                     error_msg = "The number of reasoning rounds exceeded the limit."
                     history.append(ChatMessage(role="assistant", content=error_msg))
+                    logger.debug("Yielding max rounds error")
                     yield history
                     return error_msg
 
         except Exception as e:
-            logger.error(f"Exception in run_gradio_chat: {e}", exc_info=True)
+            logger.error("Exception in run_gradio_chat: %s", e, exc_info=True)
             error_msg = f"An error occurred: {e}"
             history.append(ChatMessage(role="assistant", content=error_msg))
+            logger.debug("Yielding error history: %s", error_msg)
             yield history
             if self.force_finish:
                 last_outputs_str = self.get_answer_based_on_unfinished_reasoning(
@@ -941,5 +986,6 @@ Generate **one summarized sentence** about "function calls' responses" with nece
                     history.append(ChatMessage(role="assistant", content="**🧠 Final Analysis:**\n" + final_answer.strip()))
                 else:
                     history.append(ChatMessage(role="assistant", content=last_outputs_str.strip()))
+                logger.debug("Yielding forced final history after error")
                 yield history
             return error_msg