lollms-client 0.13.1__tar.gz → 0.14.0__tar.gz

{lollms_client-0.13.1 → lollms_client-0.14.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 0.13.1
+Version: 0.14.0
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License: Apache Software License

{lollms_client-0.13.1 → lollms_client-0.14.0}/examples/simple_text_gen_test.py

@@ -1,4 +1,4 @@
-from lollms_client import LollmsClient, ELF_COMPLETION_FORMAT
+from lollms_client import LollmsClient
 from lollms_client.lollms_types import MSG_TYPE # For callback signature
 from ascii_colors import ASCIIColors, trace_exception
 

{lollms_client-0.13.1 → lollms_client-0.14.0}/examples/text_gen.py

@@ -1,7 +1,7 @@
 from lollms_client import LollmsClient
 
 # Initialize the LollmsClient instance
-lc = LollmsClient("http://localhost:9600")
+lc = LollmsClient("lollms")
 # Generate Text
 # response = lc.generate_text(prompt="Once upon a time", stream=False, temperature=0.5)
 # print(response)

lollms_client-0.14.0/examples/text_gen_system_prompt.py

@@ -0,0 +1,28 @@
+from lollms_client import LollmsClient
+
+# Initialize the LollmsClient instance
+lc = LollmsClient("ollama",model_name="mistral-nemo:latest")
+# Generate Text
+# response = lc.generate_text(prompt="Once upon a time", stream=False, temperature=0.5)
+# print(response)
+
+# # Generate Completion
+# response = lc.generate_completion(prompt="What is the capital of France", stream=False, temperature=0.5)
+# print(response)
+
+def cb(chunk, type):
+    print(chunk,end="",flush=True)
+    
+response = lc.generate_text(prompt="One plus one equals ", system_prompt="You are a playful dude who never really answers questions correctly. always answer with quirky style.", stream=False, temperature=0.5, streaming_callback=cb)
+print()
+print(response)
+print()
+
+
+# List Mounted Personalities
+response = lc.listMountedPersonalities()
+print(response)
+
+# List Models
+response = lc.listModels()
+print(response)

{lollms_client-0.13.1 → lollms_client-0.14.0}/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__ = "0.13.1"
+__version__ = "0.14.0"
 
 # Optionally, you could define __all__ if you want to be explicit about exports
 __all__ = [

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/llm_bindings/lollms/__init__.py

@@ -49,6 +49,7 @@ class LollmsLLMBinding(LollmsLLMBinding):
     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,
@@ -106,7 +107,7 @@ class LollmsLLMBinding(LollmsLLMBinding):
         
         # Prepare request data
         data = {
-            "prompt": prompt,
+            "prompt":"!@>system: "+system_prompt+"\n"+"!@>user: "+prompt if system_prompt else prompt,
             "model_name": self.model_name,
             "personality": self.personality,
             "n_predict": n_predict,

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/llm_bindings/ollama/__init__.py

@@ -10,10 +10,11 @@ from typing import Optional, Callable, List, Union, Dict
 
 from ascii_colors import ASCIIColors, trace_exception
 import pipmaster as pm
-pm.ensure_packages(["ollama","pillow"])
+pm.ensure_packages(["ollama","pillow","tiktoken"])
 
 
 import ollama
+import tiktoken
 BindingName = "OllamaBinding"
 
 
@@ -112,9 +113,9 @@ class OllamaBinding(LollmsLLMBinding):
     def generate_text(self, 
                      prompt: str,
                      images: Optional[List[str]] = None, # List of image file paths
+                     system_prompt: str = "",
                      n_predict: Optional[int] = None,
                      stream: bool = False,
-                     system_prompt = '',
                      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
@@ -200,15 +201,16 @@ class OllamaBinding(LollmsLLMBinding):
                     )
                     return response_dict.get('message', {}).get('content', '')
             else: # Text-only
+                messages = [{'role': 'system', 'content':system_prompt},{'role': 'user', 'content': prompt}]
                 if stream:
-                    response_stream = self.ollama_client.generate(
+                    response_stream = self.ollama_client.chat(
                         model=self.model_name,
-                        prompt=prompt,
+                        messages=messages,
                         stream=True,
                         options=options if options else None
                     )
                     for chunk_dict in response_stream:
-                        chunk_content = chunk_dict.get('response', '')
+                        chunk_content = chunk_dict.message.content
                         if chunk_content:
                             full_response_text += chunk_content
                             if streaming_callback:
@@ -216,13 +218,13 @@ class OllamaBinding(LollmsLLMBinding):
                                     break
                     return full_response_text
                 else: # Not streaming
-                    response_dict = self.ollama_client.generate(
+                    response_dict = self.ollama_client.chat(
                         model=self.model_name,
-                        prompt=prompt,
+                        messages=messages,
                         stream=False,
                         options=options if options else None
                     )
-                    return response_dict.get('response', '')
+                    return response_dict.message.content
         except ollama.ResponseError as e:
             error_message = f"Ollama API ResponseError: {e.error or 'Unknown error'} (status code: {e.status_code})"
             ASCIIColors.error(error_message)
@@ -236,53 +238,31 @@ class OllamaBinding(LollmsLLMBinding):
             trace_exception(ex)
             return {"status": False, "error": error_message}
     
-    def tokenize(self, text: str) -> List[Union[int, str]]:
+    def tokenize(self, text: str) -> list:
         """
-        Tokenize the input text. For Ollama, this is complex as tokenization is model-specific
-        and best done by the server. This method provides a basic character-level tokenization
-        as a fallback or placeholder, or one could attempt to call /api/tokenize if desired.
-        The `count_tokens` method is more accurate for Ollama.
+        Tokenize the input text into a list of characters.
 
         Args:
             text (str): The text to tokenize.
 
         Returns:
-            list: List of tokens (characters or token IDs if /api/tokenize is used).
+            list: List of individual characters.
         """
-        # Basic character-level tokenization
-        # return list(text)
-
-        # For actual token IDs (slower, makes a network request):
-        api_url = f"{self.host_address.rstrip('/')}/api/tokenize"
-        payload = {"model": self.model_name, "prompt": text}
-        try:
-            response = requests.post(api_url, json=payload, timeout=10, verify=self.verify_ssl_certificate, headers=self.ollama_client_headers)
-            response.raise_for_status()
-            return response.json().get("tokens", [])
-        except Exception as e:
-            ASCIIColors.warning(f"Failed to tokenize text with Ollama server, falling back to char tokens: {e}")
-            return list(text)
-
-    def detokenize(self, tokens: List[Union[int,str]]) -> str:
+        ## Since ollama has no endpoints to tokenize the text, we use tiktoken to have a rough estimate
+        return tiktoken.model.encoding_for_model("gpt-3.5-turbo").encode(text)
+            
+    def detokenize(self, tokens: list) -> str:
         """
-        Convert a list of tokens back to text. If tokens are characters, joins them.
-        If tokens are IDs, this is non-trivial without the model's tokenizer.
+        Convert a list of tokens back to text.
 
         Args:
-            tokens (list): List of tokens to detokenize.
+            tokens (list): List of tokens (characters) to detokenize.
 
         Returns:
             str: Detokenized text.
         """
-        if not tokens:
-            return ""
-        if isinstance(tokens[0], str): # Assuming character tokens
-            return "".join(tokens)
-        else: 
-            # Detokenizing IDs from Ollama is not straightforward client-side without specific tokenizer.
-            # This is a placeholder. For Ollama, detokenization usually happens server-side.
-            ASCIIColors.warning("Detokenizing integer tokens is not accurately supported by this Ollama client binding. Returning joined string of token IDs.")
-            return "".join(map(str, tokens))
+        ## Since ollama has no endpoints to tokenize the text, we use tiktoken to have a rough estimate
+        return tiktoken.model.encoding_for_model("gpt-3.5-turbo").decode(tokens)
     
     def count_tokens(self, text: str) -> int:
         """
@@ -297,8 +277,8 @@ class OllamaBinding(LollmsLLMBinding):
         if not self.model_name:
             ASCIIColors.warning("Cannot count tokens, model_name is not set.")
             return -1
-        return count_tokens_ollama(text, self.model_name, self.ollama_client)
-    
+        #return count_tokens_ollama(text, self.model_name, self.ollama_client)
+        return len(self.tokenize(text))
     def embed(self, text: str, **kwargs) -> List[float]:
         """
         Get embeddings for the input text using Ollama API.

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/llm_bindings/openai/__init__.py

@@ -58,6 +58,7 @@ class OpenAIBinding(LollmsLLMBinding):
     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,
@@ -98,6 +99,11 @@ class OpenAIBinding(LollmsLLMBinding):
         if images:
             messages = [
                 {
+                    "role": "system",
+                    "content": system_prompt,
+                },
+
+                {                    
                     "role": "user", 
                     "content": [
                         {

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/llm_bindings/tensor_rt/__init__.py

@@ -331,6 +331,7 @@ class VLLMBinding(LollmsLLMBinding):
     def generate_text(self, 
                      prompt: str,
                      images: Optional[List[str]] = None, 
+                     system_prompt: str = "",
                      n_predict: Optional[int] = 1024, 
                      stream: bool = False, # vLLM's generate is blocking, stream is pseudo
                      temperature: float = 0.7,
@@ -381,7 +382,7 @@ class VLLMBinding(LollmsLLMBinding):
                 # If providing multi_modal_data, usually prompt_token_ids are also needed.
                 # This can get complex as it depends on how the model expects images to be interleaved.
                 # For a simple case where image comes first:
-                encoded_prompt_ids = self.tokenizer.encode(prompt)
+                encoded_prompt_ids = self.tokenizer.encode(system_prompt+"\n"+prompt if system_prompt else prompt)
                 gen_kwargs["prompt_token_ids"] = [encoded_prompt_ids] # List of lists
                 gen_kwargs["multi_modal_data"] = [{"image": mm_data_content}] # List of dicts
                 gen_kwargs["prompts"] = None # Don't use prompts if prompt_token_ids is used
@@ -389,7 +390,7 @@ class VLLMBinding(LollmsLLMBinding):
             except Exception as e_mm:
                 return {"status": False, "error": f"Multimodal prep error: {e_mm}"}
         else:
-            gen_kwargs["prompts"] = [prompt]
+            gen_kwargs["prompts"] = [system_prompt+"\n"+prompt if system_prompt else prompt]
 
         try:
             outputs = self.llm_engine.generate(**gen_kwargs, sampling_params=sampling_params)

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/llm_bindings/transformers/__init__.py

@@ -112,6 +112,7 @@ class TransformersBinding(LollmsLLMBinding):
     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,
@@ -123,8 +124,7 @@ class TransformersBinding(LollmsLLMBinding):
                       n_threads: int = 8,
                       ctx_size: int | None = None,
                       streaming_callback: Optional[Callable[[str, str], None]] = None,
-                      return_legacy_cache: bool = False,
-                      system_prompt: str = "You are a helpful assistant.") -> Union[str, dict]:
+                      return_legacy_cache: bool = False) -> Union[str, dict]:
         """
         Generate text using the Transformers model, with optional image support.
 

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/llm_bindings/vllm/__init__.py

@@ -331,6 +331,7 @@ class VLLMBinding(LollmsLLMBinding):
     def generate_text(self, 
                      prompt: str,
                      images: Optional[List[str]] = None, 
+                     system_prompt: str = "",
                      n_predict: Optional[int] = 1024, 
                      stream: bool = False, # vLLM's generate is blocking, stream is pseudo
                      temperature: float = 0.7,
@@ -381,7 +382,7 @@ class VLLMBinding(LollmsLLMBinding):
                 # If providing multi_modal_data, usually prompt_token_ids are also needed.
                 # This can get complex as it depends on how the model expects images to be interleaved.
                 # For a simple case where image comes first:
-                encoded_prompt_ids = self.tokenizer.encode(prompt)
+                encoded_prompt_ids = self.tokenizer.encode(system_prompt+"\n"+prompt if system_prompt else prompt)
                 gen_kwargs["prompt_token_ids"] = [encoded_prompt_ids] # List of lists
                 gen_kwargs["multi_modal_data"] = [{"image": mm_data_content}] # List of dicts
                 gen_kwargs["prompts"] = None # Don't use prompts if prompt_token_ids is used
@@ -389,7 +390,7 @@ class VLLMBinding(LollmsLLMBinding):
             except Exception as e_mm:
                 return {"status": False, "error": f"Multimodal prep error: {e_mm}"}
         else:
-            gen_kwargs["prompts"] = [prompt]
+            gen_kwargs["prompts"] = [system_prompt+"\n"+prompt if system_prompt else prompt]
 
         try:
             outputs = self.llm_engine.generate(**gen_kwargs, sampling_params=sampling_params)

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/lollms_core.py

@@ -329,6 +329,7 @@ class LollmsClient():
     def generate_text(self,
                      prompt: str,
                      images: Optional[List[str]] = None,
+                     system_prompt: str = "",
                      n_predict: Optional[int] = None,
                      stream: Optional[bool] = None,
                      temperature: Optional[float] = None,
@@ -365,6 +366,7 @@ class LollmsClient():
             return self.binding.generate_text(
                 prompt=prompt,
                 images=images,
+                system_prompt=system_prompt,
                 n_predict=n_predict if n_predict is not None else self.default_n_predict,
                 stream=stream if stream is not None else self.default_stream,
                 temperature=temperature if temperature is not None else self.default_temperature,

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client/lollms_llm_binding.py

@@ -32,6 +32,7 @@ class LollmsLLMBinding(ABC):
     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,

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 0.13.1
+Version: 0.14.0
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License: Apache Software License

{lollms_client-0.13.1 → lollms_client-0.14.0}/lollms_client.egg-info/SOURCES.txt

@@ -7,6 +7,7 @@ examples/text_2_audio.py
 examples/text_2_image.py
 examples/text_and_image_2_audio.py
 examples/text_gen.py
+examples/text_gen_system_prompt.py
 examples/article_summary/article_summary.py
 examples/deep_analyze/deep_analyse.py
 examples/deep_analyze/deep_analyze_multiple_files.py