praisonaiagents 0.0.117__py3-none-any.whl → 0.0.118__py3-none-any.whl

praisonaiagents/llm/llm.py

@@ -813,99 +813,56 @@ class LLM:
                         # Make one more call to get the final summary response
                         # Special handling for Ollama models that don't automatically process tool results
                         ollama_handled = False
-                        if self.model and self.model.startswith("ollama/") and tool_results:
-                            # For Ollama models, we need to explicitly ask the model to process the tool results
-                            # First, check if the response is just a JSON tool call
-                            try:
-                                # If the response_text is a valid JSON that looks like a tool call,
-                                # we need to make a follow-up call to process the results
-                                json_response = json.loads(response_text.strip())
-                                if ('name' in json_response or 'function' in json_response) and not any(word in response_text.lower() for word in ['summary', 'option', 'result', 'found']):
-                                    logging.debug("Detected Ollama returning only tool call JSON, making follow-up call to process results")
-                                    
-                                    # Create a prompt that asks the model to process the tool results based on original context
-                                    # Extract the original user query from messages
-                                    original_query = ""
-                                    for msg in reversed(messages):  # Look from the end to find the most recent user message
-                                        if msg.get("role") == "user":
-                                            content = msg.get("content", "")
-                                            # Handle list content (multimodal)
-                                            if isinstance(content, list):
-                                                for item in content:
-                                                    if isinstance(item, dict) and item.get("type") == "text":
-                                                        original_query = item.get("text", "")
-                                                        break
-                                            else:
-                                                original_query = content
-                                            if original_query:
-                                                break
-                                    
-                                    # Create a shorter follow-up prompt with all tool results
-                                    # If there's only one result, use it directly; otherwise combine them
-                                    if len(tool_results) == 1:
-                                        results_text = json.dumps(tool_results[0], indent=2)
-                                    else:
-                                        results_text = json.dumps(tool_results, indent=2)
-                                    
-                                    follow_up_prompt = f"Results:\n{results_text}\nProvide Answer to this Original Question based on the above results: '{original_query}'"
-                                    logging.debug(f"[OLLAMA_DEBUG] Original query extracted: {original_query}")
-                                    logging.debug(f"[OLLAMA_DEBUG] Follow-up prompt: {follow_up_prompt[:200]}...")
-                                    
-                                    # Make a follow-up call to process the results
-                                    follow_up_messages = [
-                                        {"role": "user", "content": follow_up_prompt}
-                                    ]
-                                    
-                                    # Get response with streaming
-                                    if verbose:
-                                        with Live(display_generating("", start_time), console=console, refresh_per_second=4) as live:
-                                            response_text = ""
-                                            for chunk in litellm.completion(
-                                                **self._build_completion_params(
-                                                    messages=follow_up_messages,
-                                                    temperature=temperature,
-                                                    stream=stream
-                                                )
-                                            ):
-                                                if chunk and chunk.choices and chunk.choices[0].delta.content:
-                                                    content = chunk.choices[0].delta.content
-                                                    response_text += content
-                                                    live.update(display_generating(response_text, start_time))
-                                    else:
-                                        response_text = ""
-                                        for chunk in litellm.completion(
-                                            **self._build_completion_params(
-                                                messages=follow_up_messages,
-                                                temperature=temperature,
-                                                stream=stream
-                                            )
-                                        ):
-                                            if chunk and chunk.choices and chunk.choices[0].delta.content:
-                                                response_text += chunk.choices[0].delta.content
-                                    
-                                    # Set flag to indicate Ollama was handled
-                                    ollama_handled = True
-                                    final_response_text = response_text.strip()
-                                    logging.debug(f"[OLLAMA_DEBUG] Ollama follow-up response: {final_response_text[:200]}...")
-                                    
-                                    # Display the response if we got one
-                                    if final_response_text and verbose:
-                                        display_interaction(
-                                            original_prompt,
-                                            final_response_text,
-                                            markdown=markdown,
-                                            generation_time=time.time() - start_time,
-                                            console=console
+                        ollama_params = self._handle_ollama_model(response_text, tool_results, messages, original_prompt)
+                        
+                        if ollama_params:
+                            # Get response with streaming
+                            if verbose:
+                                with Live(display_generating("", start_time), console=console, refresh_per_second=4) as live:
+                                    response_text = ""
+                                    for chunk in litellm.completion(
+                                        **self._build_completion_params(
+                                            messages=ollama_params["follow_up_messages"],
+                                            temperature=temperature,
+                                            stream=stream
                                         )
-                                    
-                                    # Return the final response after processing Ollama's follow-up
-                                    if final_response_text:
-                                        return final_response_text
-                                    else:
-                                        logging.warning("[OLLAMA_DEBUG] Ollama follow-up returned empty response")
-                            except (json.JSONDecodeError, KeyError):
-                                # Not a JSON response or not a tool call format, continue normally
-                                pass
+                                    ):
+                                        if chunk and chunk.choices and chunk.choices[0].delta.content:
+                                            content = chunk.choices[0].delta.content
+                                            response_text += content
+                                            live.update(display_generating(response_text, start_time))
+                            else:
+                                response_text = ""
+                                for chunk in litellm.completion(
+                                    **self._build_completion_params(
+                                        messages=ollama_params["follow_up_messages"],
+                                        temperature=temperature,
+                                        stream=stream
+                                    )
+                                ):
+                                    if chunk and chunk.choices and chunk.choices[0].delta.content:
+                                        response_text += chunk.choices[0].delta.content
+                            
+                            # Set flag to indicate Ollama was handled
+                            ollama_handled = True
+                            final_response_text = response_text.strip()
+                            logging.debug(f"[OLLAMA_DEBUG] Ollama follow-up response: {final_response_text[:200]}...")
+                            
+                            # Display the response if we got one
+                            if final_response_text and verbose:
+                                display_interaction(
+                                    ollama_params["original_prompt"],
+                                    final_response_text,
+                                    markdown=markdown,
+                                    generation_time=time.time() - start_time,
+                                    console=console
+                                )
+                            
+                            # Return the final response after processing Ollama's follow-up
+                            if final_response_text:
+                                return final_response_text
+                            else:
+                                logging.warning("[OLLAMA_DEBUG] Ollama follow-up returned empty response")
                         
                         # If reasoning_steps is True and we haven't handled Ollama already, do a single non-streaming call
                         if reasoning_steps and not ollama_handled:
@@ -1480,99 +1437,56 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                     
                     # Special handling for Ollama models that don't automatically process tool results
                     ollama_handled = False
-                    if self._is_ollama_provider() and tool_results:
-                        # For Ollama models, we need to explicitly ask the model to process the tool results
-                        # First, check if the response is just a JSON tool call
-                        try:
-                            # If the response_text is a valid JSON that looks like a tool call,
-                            # we need to make a follow-up call to process the results
-                            json_response = json.loads(response_text.strip())
-                            if ('name' in json_response or 'function' in json_response) and not any(word in response_text.lower() for word in ['summary', 'option', 'result', 'found']):
-                                logging.debug("Detected Ollama returning only tool call JSON in async mode, making follow-up call to process results")
-                                
-                                # Create a prompt that asks the model to process the tool results based on original context
-                                # Extract the original user query from messages
-                                original_query = ""
-                                for msg in reversed(messages):  # Look from the end to find the most recent user message
-                                    if msg.get("role") == "user":
-                                        content = msg.get("content", "")
-                                        # Handle list content (multimodal)
-                                        if isinstance(content, list):
-                                            for item in content:
-                                                if isinstance(item, dict) and item.get("type") == "text":
-                                                    original_query = item.get("text", "")
-                                                    break
-                                        else:
-                                            original_query = content
-                                        if original_query:
-                                            break
-                                
-                                # Create a shorter follow-up prompt with all tool results
-                                # If there's only one result, use it directly; otherwise combine them
-                                if len(tool_results) == 1:
-                                    results_text = json.dumps(tool_results[0], indent=2)
-                                else:
-                                    results_text = json.dumps(tool_results, indent=2)
-                                
-                                follow_up_prompt = f"Results:\n{results_text}\nProvide Answer to this Original Question based on the above results: '{original_query}'"
-                                logging.debug(f"[OLLAMA_DEBUG] Original query extracted: {original_query}")
-                                logging.debug(f"[OLLAMA_DEBUG] Follow-up prompt: {follow_up_prompt[:200]}...")
-                                
-                                # Make a follow-up call to process the results
-                                follow_up_messages = [
-                                    {"role": "user", "content": follow_up_prompt}
-                                ]
-                                
-                                # Get response with streaming
-                                if verbose:
-                                    response_text = ""
-                                    async for chunk in await litellm.acompletion(
-                                        **self._build_completion_params(
-                                            messages=follow_up_messages,
-                                            temperature=temperature,
-                                            stream=stream
-                                        )
-                                    ):
-                                        if chunk and chunk.choices and chunk.choices[0].delta.content:
-                                            content = chunk.choices[0].delta.content
-                                            response_text += content
-                                            print("\033[K", end="\r")
-                                            print(f"Processing results... {time.time() - start_time:.1f}s", end="\r")
-                                else:
-                                    response_text = ""
-                                    async for chunk in await litellm.acompletion(
-                                        **self._build_completion_params(
-                                            messages=follow_up_messages,
-                                            temperature=temperature,
-                                            stream=stream
-                                        )
-                                    ):
-                                        if chunk and chunk.choices and chunk.choices[0].delta.content:
-                                            response_text += chunk.choices[0].delta.content
-                                
-                                # Set flag to indicate Ollama was handled
-                                ollama_handled = True
-                                final_response_text = response_text.strip()
-                                logging.debug(f"[OLLAMA_DEBUG] Ollama follow-up response: {final_response_text[:200]}...")
-                                
-                                # Display the response if we got one
-                                if final_response_text and verbose:
-                                    display_interaction(
-                                        original_prompt,
-                                        final_response_text,
-                                        markdown=markdown,
-                                        generation_time=time.time() - start_time,
-                                        console=console
-                                    )
-                                
-                                # Return the final response after processing Ollama's follow-up
-                                if final_response_text:
-                                    return final_response_text
-                                else:
-                                    logging.warning("[OLLAMA_DEBUG] Ollama follow-up returned empty response")
-                        except (json.JSONDecodeError, KeyError):
-                            # Not a JSON response or not a tool call format, continue normally
-                            pass
+                    ollama_params = self._handle_ollama_model(response_text, tool_results, messages, original_prompt)
+                    
+                    if ollama_params:
+                        # Get response with streaming
+                        if verbose:
+                            response_text = ""
+                            async for chunk in await litellm.acompletion(
+                                **self._build_completion_params(
+                                    messages=ollama_params["follow_up_messages"],
+                                    temperature=temperature,
+                                    stream=stream
+                                )
+                            ):
+                                if chunk and chunk.choices and chunk.choices[0].delta.content:
+                                    content = chunk.choices[0].delta.content
+                                    response_text += content
+                                    print("\033[K", end="\r")
+                                    print(f"Processing results... {time.time() - start_time:.1f}s", end="\r")
+                        else:
+                            response_text = ""
+                            async for chunk in await litellm.acompletion(
+                                **self._build_completion_params(
+                                    messages=ollama_params["follow_up_messages"],
+                                    temperature=temperature,
+                                    stream=stream
+                                )
+                            ):
+                                if chunk and chunk.choices and chunk.choices[0].delta.content:
+                                    response_text += chunk.choices[0].delta.content
+                        
+                        # Set flag to indicate Ollama was handled
+                        ollama_handled = True
+                        final_response_text = response_text.strip()
+                        logging.debug(f"[OLLAMA_DEBUG] Ollama follow-up response: {final_response_text[:200]}...")
+                        
+                        # Display the response if we got one
+                        if final_response_text and verbose:
+                            display_interaction(
+                                ollama_params["original_prompt"],
+                                final_response_text,
+                                markdown=markdown,
+                                generation_time=time.time() - start_time,
+                                console=console
+                            )
+                        
+                        # Return the final response after processing Ollama's follow-up
+                        if final_response_text:
+                            return final_response_text
+                        else:
+                            logging.warning("[OLLAMA_DEBUG] Ollama follow-up returned empty response")
                     
                     # If no special handling was needed or if it's not an Ollama model
                     if reasoning_steps and not ollama_handled:
@@ -1839,6 +1753,68 @@ Output MUST be JSON with 'reflection' and 'satisfactory'.
                 
         litellm.callbacks = events
 
+    def _handle_ollama_model(self, response_text: str, tool_results: List[Any], messages: List[Dict], original_prompt: Union[str, List[Dict]]) -> Optional[Dict[str, Any]]:
+        """
+        Handle special Ollama model requirements when processing tool results.
+        
+        Args:
+            response_text: The initial response text from the model
+            tool_results: List of tool execution results
+            messages: The conversation messages list
+            original_prompt: The original user prompt
+            
+        Returns:
+            Dict with follow-up parameters if Ollama needs special handling, None otherwise
+        """
+        if not self._is_ollama_provider() or not tool_results:
+            return None
+            
+        # Check if the response is just a JSON tool call
+        try:
+            json_response = json.loads(response_text.strip())
+            if not (('name' in json_response or 'function' in json_response) and 
+                    not any(word in response_text.lower() for word in ['summary', 'option', 'result', 'found'])):
+                return None
+                
+            logging.debug("Detected Ollama returning only tool call JSON, preparing follow-up call to process results")
+            
+            # Extract the original user query from messages
+            original_query = ""
+            for msg in reversed(messages):  # Look from the end to find the most recent user message
+                if msg.get("role") == "user":
+                    content = msg.get("content", "")
+                    # Handle list content (multimodal)
+                    if isinstance(content, list):
+                        for item in content:
+                            if isinstance(item, dict) and item.get("type") == "text":
+                                original_query = item.get("text", "")
+                                break
+                    else:
+                        original_query = content
+                    if original_query:
+                        break
+            
+            # Create a shorter follow-up prompt with all tool results
+            # If there's only one result, use it directly; otherwise combine them
+            if len(tool_results) == 1:
+                results_text = json.dumps(tool_results[0], indent=2)
+            else:
+                results_text = json.dumps(tool_results, indent=2)
+            
+            follow_up_prompt = f"Results:\n{results_text}\nProvide Answer to this Original Question based on the above results: '{original_query}'"
+            logging.debug(f"[OLLAMA_DEBUG] Original query extracted: {original_query}")
+            logging.debug(f"[OLLAMA_DEBUG] Follow-up prompt: {follow_up_prompt[:200]}...")
+            
+            # Return parameters for follow-up call
+            return {
+                "follow_up_messages": [{"role": "user", "content": follow_up_prompt}],
+                "original_prompt": original_prompt
+            }
+            
+        except (json.JSONDecodeError, KeyError):
+            # Not a JSON response or not a tool call format
+            return None
+
     def _build_completion_params(self, **override_params) -> Dict[str, Any]:
         """Build parameters for litellm completion calls with all necessary config"""
         params = {

{praisonaiagents-0.0.117.dist-info → praisonaiagents-0.0.118.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: praisonaiagents
-Version: 0.0.117
+Version: 0.0.118
 Summary: Praison AI agents for completing complex tasks with Self Reflection Agents
 Author: Mervin Praison
 Requires-Python: >=3.10

{praisonaiagents-0.0.117.dist-info → praisonaiagents-0.0.118.dist-info}/RECORD

@@ -16,7 +16,7 @@ praisonaiagents/knowledge/__init__.py,sha256=xL1Eh-a3xsHyIcU4foOWF-JdWYIYBALJH9b
 praisonaiagents/knowledge/chunking.py,sha256=G6wyHa7_8V0_7VpnrrUXbEmUmptlT16ISJYaxmkSgmU,7678
 praisonaiagents/knowledge/knowledge.py,sha256=OKPar-XGyAp1ndmbOOdCgqFnTCqpOThYVSIZRxZyP58,15683
 praisonaiagents/llm/__init__.py,sha256=6lTeQ8jWi1-KiwjCDCmkHo2e-bRLq2dP0s5iJWqjO3s,1421
-praisonaiagents/llm/llm.py,sha256=-IS6gEHYVsGDxlJjrHpsnon8b9bTrHm5qhdklMqepdo,113618
+praisonaiagents/llm/llm.py,sha256=1AGCRpriOx_c3h_Bn6_DNzI1HHz0PM6jieq7RDysA68,109172
 praisonaiagents/llm/openai_client.py,sha256=0JvjCDHoH8I8kIt5vvObARkGdVaPWdTIv_FoEQ5EQPA,48973
 praisonaiagents/mcp/__init__.py,sha256=ibbqe3_7XB7VrIcUcetkZiUZS1fTVvyMy_AqCSFG8qc,240
 praisonaiagents/mcp/mcp.py,sha256=-fFx4MHffnN2woLnnV7Pzx3-1SFkn2j8Gp5F5ZIwKJ0,19698
@@ -53,7 +53,7 @@ praisonaiagents/tools/xml_tools.py,sha256=iYTMBEk5l3L3ryQ1fkUnNVYK-Nnua2Kx2S0dxN
 praisonaiagents/tools/yaml_tools.py,sha256=uogAZrhXV9O7xvspAtcTfpKSQYL2nlOTvCQXN94-G9A,14215
 praisonaiagents/tools/yfinance_tools.py,sha256=s2PBj_1v7oQnOobo2fDbQBACEHl61ftG4beG6Z979ZE,8529
 praisonaiagents/tools/train/data/generatecot.py,sha256=H6bNh-E2hqL5MW6kX3hqZ05g9ETKN2-kudSjiuU_SD8,19403
-praisonaiagents-0.0.117.dist-info/METADATA,sha256=mgOfLNqTgXkHQ-hP91BvEx1g-ZaDLGJCswHO2zM2N9U,1669
-praisonaiagents-0.0.117.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-praisonaiagents-0.0.117.dist-info/top_level.txt,sha256=_HsRddrJ23iDx5TTqVUVvXG2HeHBL5voshncAMDGjtA,16
-praisonaiagents-0.0.117.dist-info/RECORD,,
+praisonaiagents-0.0.118.dist-info/METADATA,sha256=AmgmO7hID63By-em4e2hDiWT10KeiXDpTL39qpxCDQY,1669
+praisonaiagents-0.0.118.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+praisonaiagents-0.0.118.dist-info/top_level.txt,sha256=_HsRddrJ23iDx5TTqVUVvXG2HeHBL5voshncAMDGjtA,16
+praisonaiagents-0.0.118.dist-info/RECORD,,