lollms-client 0.16.0__tar.gz → 0.17.1__tar.gz

{lollms_client-0.16.0 → lollms_client-0.17.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 0.16.0
+Version: 0.17.1
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License: Apache Software License

{lollms_client-0.16.0 → lollms_client-0.17.1}/examples/simple_text_gen_with_image_test.py

@@ -10,14 +10,14 @@ from ascii_colors import ASCIIColors, trace_exception
 # MODEL_NAME = None # Server will use its default or last loaded model
 
 # Option 2: Ollama binding
-# BINDING_NAME = "ollama"
-# HOST_ADDRESS = "http://localhost:11434" # Default Ollama host
-# MODEL_NAME = "llava:latest" # Or "llama3:latest", "phi3:latest", etc. - ensure it's pulled in Ollama
-
-# Option 2: llamacpp binding
-BINDING_NAME = "llamacpp"
-MODELS_PATH = r"E:\drumber" # Change to your own models folder
-MODEL_NAME = "llava-v1.6-mistral-7b.Q3_K_XS.gguf" # Change to your vision capable model (make sure you have a mmprj file with the gguf model with the same name but without the quantization name and with mmproj- prefix (mmproj-llava-v1.6-mistral-7b.gguf))
+BINDING_NAME = "ollama"
+HOST_ADDRESS = "http://localhost:11434" # Default Ollama host
+MODEL_NAME = "llava:latest" # Or "llama3:latest", "phi3:latest", etc. - ensure it's pulled in Ollama
+
+# Option 3: llamacpp binding
+# BINDING_NAME = "llamacpp"
+# MODELS_PATH = r"E:\drumber" # Change to your own models folder
+# MODEL_NAME = "llava-v1.6-mistral-7b.Q3_K_XS.gguf" # Change to your vision capable model (make sure you have a mmprj file with the gguf model with the same name but without the quantization name and with mmproj- prefix (mmproj-llava-v1.6-mistral-7b.gguf))
 # You can also add a clip_model_path parameter to your lc_params
 img = "E:\\drumber\\1711741182996.jpg"
 # Option 3: OpenAI binding (requires OPENAI_API_KEY environment variable or service_key)

{lollms_client-0.16.0 → lollms_client-0.17.1}/examples/text_gen.py

@@ -15,7 +15,7 @@ lc = LollmsClient("llamacpp", models_path=r"E:\drumber", model_name="llava-v1.6-
 def cb(chunk, type):
     print(chunk,end="",flush=True)
     
-response = lc.generate_text(prompt="One plus one equals ", stream=False, temperature=0.5, streaming_callback=cb)
+response = lc.generate_text(prompt="!@>user: Hi there\n!@>assistant: Hi there, how can I help you?!@>user: what is 1+1?\n!@>assistant: ", stream=False, temperature=0.5, streaming_callback=cb, split=True)
 print()
 print(response)
 print()

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/__init__.py

@@ -6,7 +6,7 @@ from lollms_client.lollms_discussion import LollmsDiscussion, LollmsMessage
 from lollms_client.lollms_utilities import PromptReshaper # Keep general utilities
 from lollms_client.lollms_functions import FunctionCalling_Library
 
-__version__ = "0.16.0"
+__version__ = "0.17.1"
 
 # Optionally, you could define __all__ if you want to be explicit about exports
 __all__ = [

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/llamacpp/__init__.py

@@ -475,7 +475,12 @@ class LlamaCppServerBinding(LollmsLLMBinding):
                                    temperature: float = 0.7, top_k: int = 40, top_p: float = 0.9,
                                    repeat_penalty: float = 1.1, repeat_last_n: Optional[int] = 64,
                                    seed: Optional[int] = None, stream: bool = False, use_chat_format: bool = True,
-                                   images: Optional[List[str]] = None, **extra_params) -> Dict:
+                                   images: Optional[List[str]] = None,
+                                    split:Optional[bool]=False, # put to true if the prompt is a discussion
+                                    user_keyword:Optional[str]="!@>user:",
+                                    ai_keyword:Optional[str]="!@>assistant:", 
+                                   
+                                   **extra_params) -> Dict:
         payload_params = {
             "temperature": self.server_args.get("temperature", 0.7), "top_k": self.server_args.get("top_k", 40),
             "top_p": self.server_args.get("top_p", 0.9), "repeat_penalty": self.server_args.get("repeat_penalty", 1.1),
@@ -495,6 +500,10 @@ class LlamaCppServerBinding(LollmsLLMBinding):
             messages = []
             if system_prompt and system_prompt.strip(): messages.append({"role": "system", "content": system_prompt})
             user_content: Union[str, List[Dict[str, Any]]] = prompt
+            if split:
+                messages += self.split_discussion(user_content,user_keyword=user_keyword, ai_keyword=ai_keyword)
+            else:
+                messages.append({"role": "user", "content": user_content})
             if images and self.clip_model_path: # Use the binding's current clip_model_path
                 image_parts = []
                 for img_path in images:
@@ -503,8 +512,7 @@ class LlamaCppServerBinding(LollmsLLMBinding):
                         image_type = Path(img_path).suffix[1:].lower() or "png"; image_type = "jpeg" if image_type == "jpg" else image_type
                         image_parts.append({"type": "image_url", "image_url": {"url": f"data:image/{image_type};base64,{encoded_string}"}})
                     except Exception as ex: trace_exception(ex)
-                user_content = [{"type": "text", "text": prompt}] + image_parts # type: ignore
-            messages.append({"role": "user", "content": user_content})
+                messages[-1]["content"] =[{"type": "text", "text": messages[-1]["content"]}] +  image_parts # type: ignore
             final_payload = {"messages": messages, "stream": stream, **payload_params}
             if 'n_predict' in final_payload: final_payload['max_tokens'] = final_payload.pop('n_predict')
             return final_payload
@@ -521,16 +529,57 @@ class LlamaCppServerBinding(LollmsLLMBinding):
                 if image_data_list: final_payload["image_data"] = image_data_list
             return final_payload
 
-    def generate_text(self, prompt: str, images: Optional[List[str]] = None, system_prompt: str = "",
-                     n_predict: Optional[int] = None, stream: bool = False, temperature: float = None,
-                     top_k: int = None, top_p: float = None, repeat_penalty: float = None,
-                     repeat_last_n: Optional[int] = None, seed: Optional[int] = None,
-                     streaming_callback: Optional[Callable[[str, int], bool]] = None,
-                     use_chat_format_override: Optional[bool] = None, **generation_kwargs) -> Union[str, Dict[str, any]]:
+
+    def generate_text(self,
+                     prompt: str,
+                     images: Optional[List[str]] = None,
+                     system_prompt: str = "",
+                     n_predict: Optional[int] = None,
+                     stream: Optional[bool] = None,
+                     temperature: float = 0.7, # Ollama default is 0.8, common default 0.7
+                     top_k: int = 40,          # Ollama default is 40
+                     top_p: float = 0.9,       # Ollama default is 0.9
+                     repeat_penalty: float = 1.1, # Ollama default is 1.1
+                     repeat_last_n: int = 64,  # Ollama default is 64
+                     seed: Optional[int] = None,
+                     n_threads: Optional[int] = None,
+                     ctx_size: int | None = None,
+                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:", 
+                     **generation_kwargs
+                     ) -> Union[str, dict]:
+        """
+        Generate text using the active LLM binding, using instance defaults if parameters are not provided.
+
+        Args:
+            prompt (str): The input prompt for text generation.
+            images (Optional[List[str]]): List of image file paths for multimodal generation.
+            n_predict (Optional[int]): Maximum number of tokens to generate. Uses instance default if None.
+            stream (Optional[bool]): Whether to stream the output. Uses instance default if None.
+            temperature (Optional[float]): Sampling temperature. Uses instance default if None.
+            top_k (Optional[int]): Top-k sampling parameter. Uses instance default if None.
+            top_p (Optional[float]): Top-p sampling parameter. Uses instance default if None.
+            repeat_penalty (Optional[float]): Penalty for repeated tokens. Uses instance default if None.
+            repeat_last_n (Optional[int]): Number of previous tokens to consider for repeat penalty. Uses instance default if None.
+            seed (Optional[int]): Random seed for generation. Uses instance default if None.
+            n_threads (Optional[int]): Number of threads to use. Uses instance default if None.
+            ctx_size (int | None): Context size override for this generation.
+            streaming_callback (Optional[Callable[[str, str], None]]): Callback function for streaming output.
+                - First parameter (str): The chunk of text received.
+                - Second parameter (str): The message type (e.g., MSG_TYPE.MSG_TYPE_CHUNK).
+            split:Optional[bool]: put to true if the prompt is a discussion
+            user_keyword:Optional[str]: when splitting we use this to extract user prompt 
+            ai_keyword:Optional[str]": when splitting we use this to extract ai prompt
+
+        Returns:
+            Union[str, dict]: Generated text or error dictionary if failed.
+        """
         if not self.server_process or not self.server_process.is_healthy:
              return {"status": False, "error": "Llama.cpp server is not running or not healthy."}
 
-        _use_chat_format = use_chat_format_override if use_chat_format_override is not None else (self.default_completion_format == ELF_COMPLETION_FORMAT.Chat)
+        _use_chat_format = True
         payload = self._prepare_generation_payload(
             prompt=prompt, system_prompt=system_prompt, n_predict=n_predict,
             temperature=temperature if temperature is not None else self.server_args.get("temperature",0.7),
@@ -539,7 +588,8 @@ class LlamaCppServerBinding(LollmsLLMBinding):
             repeat_penalty=repeat_penalty if repeat_penalty is not None else self.server_args.get("repeat_penalty",1.1),
             repeat_last_n=repeat_last_n if repeat_last_n is not None else self.server_args.get("repeat_last_n",64),
             seed=seed if seed is not None else self.server_args.get("seed", -1), stream=stream,
-            use_chat_format=_use_chat_format, images=images, **generation_kwargs
+            use_chat_format=_use_chat_format, images=images,
+            split= split, user_keyword=user_keyword, ai_keyword=ai_keyword, **generation_kwargs
         )
         endpoint = "/v1/chat/completions" if _use_chat_format else "/completion"
         request_url = self._get_request_url(endpoint)

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/lollms/__init__.py

@@ -46,43 +46,50 @@ class LollmsLLMBinding(LollmsLLMBinding):
         self.personality = personality
         self.model = None
     
-    def generate_text(self, 
+    def generate_text(self,
                      prompt: str,
                      images: Optional[List[str]] = None,
                      system_prompt: str = "",
                      n_predict: Optional[int] = None,
-                     stream: bool = False,
-                     temperature: float = 0.1,
-                     top_k: int = 50,
-                     top_p: float = 0.95,
-                     repeat_penalty: float = 0.8,
-                     repeat_last_n: int = 40,
+                     stream: Optional[bool] = None,
+                     temperature: Optional[float] = None,
+                     top_k: Optional[int] = None,
+                     top_p: Optional[float] = None,
+                     repeat_penalty: Optional[float] = None,
+                     repeat_last_n: Optional[int] = None,
                      seed: Optional[int] = None,
-                     n_threads: int = 8,
+                     n_threads: Optional[int] = None,
                      ctx_size: int | None = None,
-                     streaming_callback: Optional[Callable[[str, str], None]] = None) -> Union[str, dict]:
+                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
+                     ) -> Union[str, dict]:
         """
-        Generate text using the LOLLMS service, with optional image support.
+        Generate text using the active LLM binding, using instance defaults if parameters are not provided.
 
         Args:
             prompt (str): The input prompt for text generation.
             images (Optional[List[str]]): List of image file paths for multimodal generation.
-                If provided, uses the /lollms_generate_with_images endpoint.
-            n_predict (Optional[int]): Maximum number of tokens to generate.
-            stream (bool): Whether to stream the output. Defaults to False.
-            temperature (float): Sampling temperature. Defaults to 0.1.
-            top_k (int): Top-k sampling parameter. Defaults to 50.
-            top_p (float): Top-p sampling parameter. Defaults to 0.95.
-            repeat_penalty (float): Penalty for repeated tokens. Defaults to 0.8.
-            repeat_last_n (int): Number of previous tokens to consider for repeat penalty. Defaults to 40.
-            seed (Optional[int]): Random seed for generation.
-            n_threads (int): Number of threads to use. Defaults to 8.
-            streaming_callback (Optional[Callable[[str, str], None]]): Callback for streaming output.
-                - First parameter (str): The chunk of text received from the stream.
-                - Second parameter (str): The message type (typically MSG_TYPE.MSG_TYPE_CHUNK).
+            n_predict (Optional[int]): Maximum number of tokens to generate. Uses instance default if None.
+            stream (Optional[bool]): Whether to stream the output. Uses instance default if None.
+            temperature (Optional[float]): Sampling temperature. Uses instance default if None.
+            top_k (Optional[int]): Top-k sampling parameter. Uses instance default if None.
+            top_p (Optional[float]): Top-p sampling parameter. Uses instance default if None.
+            repeat_penalty (Optional[float]): Penalty for repeated tokens. Uses instance default if None.
+            repeat_last_n (Optional[int]): Number of previous tokens to consider for repeat penalty. Uses instance default if None.
+            seed (Optional[int]): Random seed for generation. Uses instance default if None.
+            n_threads (Optional[int]): Number of threads to use. Uses instance default if None.
+            ctx_size (int | None): Context size override for this generation.
+            streaming_callback (Optional[Callable[[str, str], None]]): Callback function for streaming output.
+                - First parameter (str): The chunk of text received.
+                - Second parameter (str): The message type (e.g., MSG_TYPE.MSG_TYPE_CHUNK).
+            split:Optional[bool]: put to true if the prompt is a discussion
+            user_keyword:Optional[str]: when splitting we use this to extract user prompt 
+            ai_keyword:Optional[str]": when splitting we use this to extract ai prompt
 
         Returns:
-            Union[str, dict]: Generated text if successful, or a dictionary with status and error if failed.
+            Union[str, dict]: Generated text or error dictionary if failed.
         """
         # Determine endpoint based on presence of images
         endpoint = "/lollms_generate_with_images" if images else "/lollms_generate"

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/ollama/__init__.py

@@ -109,47 +109,53 @@ class OllamaBinding(LollmsLLMBinding):
             self.ollama_client = None # Ensure it's None if initialization fails
             # Optionally re-raise or handle so the binding is clearly unusable
             raise ConnectionError(f"Could not connect or initialize Ollama client at {self.host_address}: {e}") from e
-    
-    def generate_text(self, 
+
+    def generate_text(self,
                      prompt: str,
-                     images: Optional[List[str]] = None, # List of image file paths
+                     images: Optional[List[str]] = None,
                      system_prompt: str = "",
                      n_predict: Optional[int] = None,
-                     stream: bool = False,
+                     stream: Optional[bool] = None,
                      temperature: float = 0.7, # Ollama default is 0.8, common default 0.7
                      top_k: int = 40,          # Ollama default is 40
                      top_p: float = 0.9,       # Ollama default is 0.9
                      repeat_penalty: float = 1.1, # Ollama default is 1.1
                      repeat_last_n: int = 64,  # Ollama default is 64
                      seed: Optional[int] = None,
-                     n_threads: Optional[int] = None, # Ollama calls this num_thread
-                     ctx_size: Optional[int] = None,  # Ollama calls this num_ctx
-                     streaming_callback: Optional[Callable[[str, int], bool]] = None
-                     ) -> Union[str, Dict[str, any]]:
+                     n_threads: Optional[int] = None,
+                     ctx_size: int | None = None,
+                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
+                     ) -> Union[str, dict]:
         """
-        Generate text using the Ollama service, with optional image support.
+        Generate text using the active LLM binding, using instance defaults if parameters are not provided.
 
         Args:
             prompt (str): The input prompt for text generation.
             images (Optional[List[str]]): List of image file paths for multimodal generation.
-            n_predict (Optional[int]): Maximum number of tokens to generate (num_predict).
-            stream (bool): Whether to stream the output. Defaults to False.
-            temperature (float): Sampling temperature.
-            top_k (int): Top-k sampling parameter.
-            top_p (float): Top-p sampling parameter.
-            repeat_penalty (float): Penalty for repeated tokens.
-            repeat_last_n (int): Number of previous tokens to consider for repeat penalty.
-            seed (Optional[int]): Random seed for generation.
-            n_threads (Optional[int]): Number of threads to use (num_thread).
-            ctx_size (Optional[int]): Context window size (num_ctx).
-            streaming_callback (Optional[Callable[[str, int], bool]]): Callback for streaming output.
-                - First parameter (str): The chunk of text received from the stream.
-                - Second parameter (int): The message type (typically MSG_TYPE.MSG_TYPE_CHUNK).
-                Return False to stop streaming.
+            n_predict (Optional[int]): Maximum number of tokens to generate. Uses instance default if None.
+            stream (Optional[bool]): Whether to stream the output. Uses instance default if None.
+            temperature (Optional[float]): Sampling temperature. Uses instance default if None.
+            top_k (Optional[int]): Top-k sampling parameter. Uses instance default if None.
+            top_p (Optional[float]): Top-p sampling parameter. Uses instance default if None.
+            repeat_penalty (Optional[float]): Penalty for repeated tokens. Uses instance default if None.
+            repeat_last_n (Optional[int]): Number of previous tokens to consider for repeat penalty. Uses instance default if None.
+            seed (Optional[int]): Random seed for generation. Uses instance default if None.
+            n_threads (Optional[int]): Number of threads to use. Uses instance default if None.
+            ctx_size (int | None): Context size override for this generation.
+            streaming_callback (Optional[Callable[[str, str], None]]): Callback function for streaming output.
+                - First parameter (str): The chunk of text received.
+                - Second parameter (str): The message type (e.g., MSG_TYPE.MSG_TYPE_CHUNK).
+            split:Optional[bool]: put to true if the prompt is a discussion
+            user_keyword:Optional[str]: when splitting we use this to extract user prompt 
+            ai_keyword:Optional[str]": when splitting we use this to extract ai prompt
 
         Returns:
-            Union[str, Dict[str, any]]: Generated text if successful, or a dictionary with status and error if failed.
+            Union[str, dict]: Generated text or error dictionary if failed.
         """
+
         if not self.ollama_client:
              return {"status": False, "error": "Ollama client not initialized."}
 
@@ -175,8 +181,15 @@ class OllamaBinding(LollmsLLMBinding):
                     # If images were base64 strings, they would need decoding to bytes first.
                     processed_images.append(img_path)
 
-                messages = [{'role': 'system', 'content':system_prompt},{'role': 'user', 'content': prompt, 'images': processed_images if processed_images else None}]
-                
+                messages = [
+                            {'role': 'system', 'content':system_prompt},
+                        ]
+                if split:
+                    messages += self.split_discussion(prompt,user_keyword=user_keyword, ai_keyword=ai_keyword)
+                    if processed_images:
+                        messages[-1]["images"]=processed_images
+                else:
+                    messages.append({'role': 'user', 'content': prompt, 'images': processed_images if processed_images else None})
                 if stream:
                     response_stream = self.ollama_client.chat(
                         model=self.model_name,
@@ -201,7 +214,14 @@ class OllamaBinding(LollmsLLMBinding):
                     )
                     return response_dict.get('message', {}).get('content', '')
             else: # Text-only
-                messages = [{'role': 'system', 'content':system_prompt},{'role': 'user', 'content': prompt}]
+                messages = [
+                            {'role': 'system', 'content':system_prompt},
+                        ]
+                if split:
+                    messages += self.split_discussion(prompt,user_keyword=user_keyword, ai_keyword=ai_keyword)
+                else:
+                    messages.append({'role': 'user', 'content': prompt})
+
                 if stream:
                     response_stream = self.ollama_client.chat(
                         model=self.model_name,

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/openai/__init__.py

@@ -55,42 +55,50 @@ class OpenAIBinding(LollmsLLMBinding):
         self.completion_format = ELF_COMPLETION_FORMAT.Chat
 
     
-    def generate_text(self, 
-                    prompt: str,
-                    images: Optional[List[str]] = None,
-                    system_prompt: str = "",
-                    n_predict: Optional[int] = None,
-                    stream: bool = False,
-                    temperature: float = 0.1,
-                    top_k: int = 50,
-                    top_p: float = 0.95,
-                    repeat_penalty: float = 0.8,
-                    repeat_last_n: int = 40,
-                    seed: Optional[int] = None,
-                    n_threads: int = 8,
-                    ctx_size: int | None = None,
-                    streaming_callback: Optional[Callable[[str, str], None]] = None) -> str:
+    def generate_text(self,
+                     prompt: str,
+                     images: Optional[List[str]] = None,
+                     system_prompt: str = "",
+                     n_predict: Optional[int] = None,
+                     stream: Optional[bool] = None,
+                     temperature: float = 0.7, 
+                     top_k: int = 40,          
+                     top_p: float = 0.9,       
+                     repeat_penalty: float = 1.1, 
+                     repeat_last_n: int = 64, 
+                     seed: Optional[int] = None,
+                     n_threads: Optional[int] = None,
+                     ctx_size: int | None = None,
+                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
+                     ) -> Union[str, dict]:
         """
-        Generate text based on the provided prompt and parameters.
+        Generate text using the active LLM binding, using instance defaults if parameters are not provided.
 
         Args:
             prompt (str): The input prompt for text generation.
             images (Optional[List[str]]): List of image file paths for multimodal generation.
-            n_predict (Optional[int]): Maximum number of tokens to generate.
-            stream (bool): Whether to stream the output. Defaults to False.
-            temperature (float): Sampling temperature. Defaults to 0.1.
-            top_k (int): Top-k sampling parameter. Defaults to 50.
-            top_p (float): Top-p sampling parameter. Defaults to 0.95.
-            repeat_penalty (float): Penalty for repeated tokens. Defaults to 0.8.
-            repeat_last_n (int): Number of previous tokens to consider for repeat penalty. Defaults to 40.
-            seed (Optional[int]): Random seed for generation.
-            n_threads (int): Number of threads to use. Defaults to 8.
+            n_predict (Optional[int]): Maximum number of tokens to generate. Uses instance default if None.
+            stream (Optional[bool]): Whether to stream the output. Uses instance default if None.
+            temperature (Optional[float]): Sampling temperature. Uses instance default if None.
+            top_k (Optional[int]): Top-k sampling parameter. Uses instance default if None.
+            top_p (Optional[float]): Top-p sampling parameter. Uses instance default if None.
+            repeat_penalty (Optional[float]): Penalty for repeated tokens. Uses instance default if None.
+            repeat_last_n (Optional[int]): Number of previous tokens to consider for repeat penalty. Uses instance default if None.
+            seed (Optional[int]): Random seed for generation. Uses instance default if None.
+            n_threads (Optional[int]): Number of threads to use. Uses instance default if None.
+            ctx_size (int | None): Context size override for this generation.
             streaming_callback (Optional[Callable[[str, str], None]]): Callback function for streaming output.
                 - First parameter (str): The chunk of text received.
                 - Second parameter (str): The message type (e.g., MSG_TYPE.MSG_TYPE_CHUNK).
+            split:Optional[bool]: put to true if the prompt is a discussion
+            user_keyword:Optional[str]: when splitting we use this to extract user prompt 
+            ai_keyword:Optional[str]": when splitting we use this to extract ai prompt
 
         Returns:
-            str: Generated text or error dictionary if failed.
+            Union[str, dict]: Generated text or error dictionary if failed.
         """
         count = 0
         output = ""
@@ -101,16 +109,17 @@ class OpenAIBinding(LollmsLLMBinding):
                 {
                     "role": "system",
                     "content": system_prompt,
-                },
-
-                {                    
-                    "role": "user", 
-                    "content": [
+                }
+            ]
+            if split:
+                messages += self.split_discussion(prompt,user_keyword=user_keyword, ai_keyword=ai_keyword)
+                if images:
+                    messages[-1]["content"] = [
                         {
                             "type": "text",
-                            "text": prompt
+                            "text": messages[-1]["content"]
                         }
-                    ] + [
+                    ]+[
                         {
                             "type": "image_url",
                             "image_url": {
@@ -119,8 +128,26 @@ class OpenAIBinding(LollmsLLMBinding):
                         }
                         for image_path in images
                     ]
-                }
-            ]
+            else:
+                messages.append({
+                        'role': 'user', 
+                        'content': [
+                                        {
+                                            "type": "text",
+                                            "text": prompt
+                                        }
+                                    ] + [
+                                        {
+                                            "type": "image_url",
+                                            "image_url": {
+                                                "url": f"data:image/jpeg;base64,{encode_image(image_path)}"
+                                            }
+                                        }
+                                        for image_path in images
+                                    ]
+                    }
+                )
+            
         else:
             messages = [{"role": "user", "content": prompt}]
 

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/openllm/__init__.py

@@ -154,7 +154,10 @@ class OpenLLMBinding(LollmsLLMBinding):
                      seed: Optional[int] = None,
                      # n_threads: Optional[int] = None, # Server-side config for OpenLLM
                      # ctx_size: Optional[int] = None,  # Server-side config, though some models might allow via llm_config
-                     streaming_callback: Optional[Callable[[str, int], bool]] = None
+                     streaming_callback: Optional[Callable[[str, int], bool]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",                     
                      ) -> Union[str, Dict[str, any]]:
         
         if not self.openllm_client:

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/pythonllamacpp/__init__.py

@@ -216,6 +216,9 @@ class PythonLlamaCppBinding(LollmsLLMBinding):
                      streaming_callback: Optional[Callable[[str, int], bool]] = None,
                      use_chat_format: bool = True, 
                      grammar: Optional[Union[str, LlamaGrammar]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
                      **generation_kwargs 
                      ) -> Union[str, Dict[str, any]]:
         

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/tensor_rt/__init__.py

@@ -341,7 +341,10 @@ class VLLMBinding(LollmsLLMBinding):
                      repeat_last_n: int = 64, # Note: vLLM applies penalty to full context
                      seed: Optional[int] = None,
                      n_threads: int = 8, # Note: vLLM manages its own threading/parallelism
-                     streaming_callback: Optional[Callable[[str, int], bool]] = None
+                     streaming_callback: Optional[Callable[[str, int], bool]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
                      ) -> Union[str, Dict[str, any]]:
         if not self.llm_engine: return {"status": False, "error": "Engine not loaded."}
 

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/transformers/__init__.py

@@ -312,6 +312,9 @@ class HuggingFaceHubBinding(LollmsLLMBinding):
                      seed: Optional[int] = None,
                      stop_words: Optional[List[str]] = None, # Added custom stop_words
                      streaming_callback: Optional[Callable[[str, int], bool]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
                      use_chat_format_override: Optional[bool] = None,
                      **generation_kwargs 
                      ) -> Union[str, Dict[str, Any]]:

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/llm_bindings/vllm/__init__.py

@@ -341,7 +341,10 @@ class VLLMBinding(LollmsLLMBinding):
                      repeat_last_n: int = 64, # Note: vLLM applies penalty to full context
                      seed: Optional[int] = None,
                      n_threads: int = 8, # Note: vLLM manages its own threading/parallelism
-                     streaming_callback: Optional[Callable[[str, int], bool]] = None
+                     streaming_callback: Optional[Callable[[str, int], bool]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
                      ) -> Union[str, Dict[str, any]]:
         if not self.llm_engine: return {"status": False, "error": "Engine not loaded."}
 

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/lollms_core.py

@@ -11,7 +11,7 @@ from lollms_client.lollms_stt_binding import LollmsSTTBinding, LollmsSTTBindingM
 from lollms_client.lollms_ttv_binding import LollmsTTVBinding, LollmsTTVBindingManager
 from lollms_client.lollms_ttm_binding import LollmsTTMBinding, LollmsTTMBindingManager
 
-import json
+import re
 from enum import Enum
 import base64
 import requests
@@ -61,11 +61,12 @@ class LollmsClient():
                  ctx_size: Optional[int] = 8192,
                  n_predict: Optional[int] = 4096,
                  stream: bool = False,
-                 temperature: float = 0.1,
-                 top_k: int = 50,
-                 top_p: float = 0.95,
-                 repeat_penalty: float = 0.8,
-                 repeat_last_n: int = 40,
+                 temperature: float = 0.7, # Ollama default is 0.8, common default 0.7
+                 top_k: int = 40,          # Ollama default is 40
+                 top_p: float = 0.9,       # Ollama default is 0.9
+                 repeat_penalty: float = 1.1, # Ollama default is 1.1
+                 repeat_last_n: int = 64,  # Ollama default is 64
+
                  seed: Optional[int] = None,
                  n_threads: int = 8,
                  streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
@@ -362,7 +363,11 @@ class LollmsClient():
                      seed: Optional[int] = None,
                      n_threads: Optional[int] = None,
                      ctx_size: int | None = None,
-                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None) -> Union[str, dict]:
+                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
+                     ) -> Union[str, dict]:
         """
         Generate text using the active LLM binding, using instance defaults if parameters are not provided.
 
@@ -380,6 +385,9 @@ class LollmsClient():
             n_threads (Optional[int]): Number of threads to use. Uses instance default if None.
             ctx_size (int | None): Context size override for this generation.
             streaming_callback (Optional[Callable[[str, MSG_TYPE], None]]): Callback for streaming output.
+            split:Optional[bool]: put to true if the prompt is a discussion
+            user_keyword:Optional[str]: when splitting we use this to extract user prompt 
+            ai_keyword:Optional[str]": when splitting we use this to extract ai prompt
 
         Returns:
             Union[str, dict]: Generated text or error dictionary if failed.
@@ -399,7 +407,10 @@ class LollmsClient():
                 seed=seed if seed is not None else self.default_seed,
                 n_threads=n_threads if n_threads is not None else self.default_n_threads,
                 ctx_size = ctx_size if ctx_size is not None else self.default_ctx_size,
-                streaming_callback=streaming_callback if streaming_callback is not None else self.default_streaming_callback
+                streaming_callback=streaming_callback if streaming_callback is not None else self.default_streaming_callback,
+                split= split,
+                user_keyword=user_keyword,
+                ai_keyword=ai_keyword
             )
         raise RuntimeError("LLM binding not initialized.")
 
@@ -981,7 +992,6 @@ Do not split the code in multiple tags.
         Ranks answers for a question from best to worst using LLM JSON generation.
         (Implementation requires self.generate_code which uses self.generate_text)
         """
-        # ... (Implementation as provided before, relies on self.generate_code) ...
         if not callback:
             callback = self.sink
 
@@ -1567,6 +1577,7 @@ Provide the final aggregated answer in {output_format} format, directly addressi
         callback("Deep analysis complete.", MSG_TYPE.MSG_TYPE_STEP_END)
         return final_output
 
+
 def chunk_text(text, tokenizer, detokenizer, chunk_size, overlap, use_separators=True):
     """
     Chunks text based on token count.
@@ -1646,3 +1657,5 @@ def chunk_text(text, tokenizer, detokenizer, chunk_size, overlap, use_separators
                 break
 
     return chunks
+
+

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/lollms_llm_binding.py

@@ -2,13 +2,14 @@
 from abc import ABC, abstractmethod
 import importlib
 from pathlib import Path
-from typing import Optional, Callable, List
+from typing import Optional, Callable, List, Union
 from lollms_client.lollms_types import ELF_COMPLETION_FORMAT
 import importlib
 from pathlib import Path
 from typing import Optional
 from ascii_colors import trace_exception
-
+from lollms_client.lollms_types import MSG_TYPE
+import re
 class LollmsLLMBinding(ABC):
     """Abstract base class for all LOLLMS LLM bindings"""
     
@@ -25,41 +26,50 @@ class LollmsLLMBinding(ABC):
         self.model_name = None #Must be set by the instance
     
     @abstractmethod
-    def generate_text(self, 
+    def generate_text(self,
                      prompt: str,
                      images: Optional[List[str]] = None,
                      system_prompt: str = "",
                      n_predict: Optional[int] = None,
-                     stream: bool = False,
-                     temperature: float = 0.1,
-                     top_k: int = 50,
-                     top_p: float = 0.95,
-                     repeat_penalty: float = 0.8,
-                     repeat_last_n: int = 40,
+                     stream: Optional[bool] = None,
+                     temperature: Optional[float] = None,
+                     top_k: Optional[int] = None,
+                     top_p: Optional[float] = None,
+                     repeat_penalty: Optional[float] = None,
+                     repeat_last_n: Optional[int] = None,
                      seed: Optional[int] = None,
-                     n_threads: int = 8,
-                     streaming_callback: Optional[Callable[[str, str], None]] = None) -> str:
+                     n_threads: Optional[int] = None,
+                     ctx_size: int | None = None,
+                     streaming_callback: Optional[Callable[[str, MSG_TYPE], None]] = None,
+                     split:Optional[bool]=False, # put to true if the prompt is a discussion
+                     user_keyword:Optional[str]="!@>user:",
+                     ai_keyword:Optional[str]="!@>assistant:",
+                     ) -> Union[str, dict]:
         """
-        Generate text based on the provided prompt and parameters.
+        Generate text using the active LLM binding, using instance defaults if parameters are not provided.
 
         Args:
             prompt (str): The input prompt for text generation.
             images (Optional[List[str]]): List of image file paths for multimodal generation.
-            n_predict (Optional[int]): Maximum number of tokens to generate.
-            stream (bool): Whether to stream the output. Defaults to False.
-            temperature (float): Sampling temperature. Defaults to 0.1.
-            top_k (int): Top-k sampling parameter. Defaults to 50.
-            top_p (float): Top-p sampling parameter. Defaults to 0.95.
-            repeat_penalty (float): Penalty for repeated tokens. Defaults to 0.8.
-            repeat_last_n (int): Number of previous tokens to consider for repeat penalty. Defaults to 40.
-            seed (Optional[int]): Random seed for generation.
-            n_threads (int): Number of threads to use. Defaults to 8.
+            n_predict (Optional[int]): Maximum number of tokens to generate. Uses instance default if None.
+            stream (Optional[bool]): Whether to stream the output. Uses instance default if None.
+            temperature (Optional[float]): Sampling temperature. Uses instance default if None.
+            top_k (Optional[int]): Top-k sampling parameter. Uses instance default if None.
+            top_p (Optional[float]): Top-p sampling parameter. Uses instance default if None.
+            repeat_penalty (Optional[float]): Penalty for repeated tokens. Uses instance default if None.
+            repeat_last_n (Optional[int]): Number of previous tokens to consider for repeat penalty. Uses instance default if None.
+            seed (Optional[int]): Random seed for generation. Uses instance default if None.
+            n_threads (Optional[int]): Number of threads to use. Uses instance default if None.
+            ctx_size (int | None): Context size override for this generation.
             streaming_callback (Optional[Callable[[str, str], None]]): Callback function for streaming output.
                 - First parameter (str): The chunk of text received.
                 - Second parameter (str): The message type (e.g., MSG_TYPE.MSG_TYPE_CHUNK).
+            split:Optional[bool]: put to true if the prompt is a discussion
+            user_keyword:Optional[str]: when splitting we use this to extract user prompt 
+            ai_keyword:Optional[str]": when splitting we use this to extract ai prompt
 
         Returns:
-            str: Generated text or error dictionary if failed.
+            Union[str, dict]: Generated text or error dictionary if failed.
         """
         pass
     
@@ -146,6 +156,52 @@ class LollmsLLMBinding(ABC):
         pass
 
 
+    def split_discussion(self, lollms_prompt_string: str, system_keyword="!@>system:", user_keyword="!@>user:", ai_keyword="!@>assistant:") -> list:
+        """
+        Splits a LoLLMs prompt into a list of OpenAI-style messages.
+        If the very first chunk has no prefix, it's assigned to "system".
+        """
+        # Regex to split on any of the three prefixes (lookahead)
+        pattern = r"(?={}|{}|{})".format(
+            re.escape(system_keyword),
+            re.escape(user_keyword),
+            re.escape(ai_keyword)
+        )
+        parts = re.split(pattern, lollms_prompt_string)
+        messages = []
+
+        for part in parts:
+            part = part.strip()
+            if not part:
+                continue
+
+            # Determine role and strip prefix if present
+            if part.startswith(system_keyword):
+                role = "system"
+                content = part[len(system_keyword):].strip()
+            elif part.startswith(user_keyword):
+                role = "user"
+                content = part[len(user_keyword):].strip()
+            elif part.startswith(ai_keyword):
+                role = "assistant"
+                content = part[len(ai_keyword):].strip()
+            else:
+                # No prefix: if it's the first valid chunk, treat as system
+                if not messages:
+                    role = "system"
+                    content = part
+                else:
+                    # otherwise skip unrecognized segments
+                    continue
+
+            messages.append({"role": role, "content": content})
+            if messages[-1]["content"]=="":
+                del messages[-1]
+        return messages
+
+
+
+
 class LollmsLLMBindingManager:
     """Manages binding discovery and instantiation"""
 

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client/lollms_utilities.py

@@ -1,10 +1,12 @@
 import urllib
 import numpy
 from pathlib import Path
-from pipmaster import PackageManager
+import pipmaster as pm
 from PIL import Image
 import io
 import base64
+import re
+import numpy as np
 class PromptReshaper:
     def __init__(self, template:str):
         self.template = template
@@ -122,8 +124,8 @@ def remove_text_from_string(string: str, text_to_find:str):
 
 
 def process_ai_output(output, images, output_folder):
-    if not PackageManager.is_installed("cv2"):
-        PackageManager.install("opencv-python")
+    if not pm.is_installed("opencv-python"):
+        pm.install("opencv-python")
     import cv2
     images = [cv2.imread(str(img)) for img in images]
     # Find all bounding box entries in the output

{lollms_client-0.16.0 → lollms_client-0.17.1}/lollms_client.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 0.16.0
+Version: 0.17.1
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License: Apache Software License