lollms-client 0.13.0__py3-none-any.whl → 0.13.2__py3-none-any.whl

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
 
@@ -97,6 +97,8 @@ def test_text_generation():
         print(emb)
 
         # else: if callback returns False early, response_stream might be partial.
+        nb_tokens = lc.count_tokens("")
+        ASCIIColors.yellow("Number of tokens of : Testing count of tokens\n"+f"{nb_tokens}")
 
         # 3. Test generation with a specific model (if applicable and different from default)
         #    This tests the switch_model or model loading mechanism of the binding.
@@ -168,4 +170,4 @@ def test_text_generation():
         trace_exception(e)
 
 if __name__ == "__main__":
-    test_text_generation()
+    test_text_generation()

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__ = "0.13.0"
+__version__ = "0.13.2"
 
 # Optionally, you could define __all__ if you want to be explicit about exports
 __all__ = [
@@ -18,4 +18,4 @@ __all__ = [
     "LollmsMessage",
     "PromptReshaper",
     "FunctionCalling_Library"
-]
+]

lollms_client/llm_bindings/ollama/__init__.py

@@ -10,72 +10,55 @@ 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"
 
 
 def count_tokens_ollama(
     text_to_tokenize: str,
     model_name: str,
-    ollama_host: str = "http://localhost:11434",
-    timeout: int = 30,
-    verify_ssl_certificate: bool = True,
-    headers: Optional[Dict[str, str]] = None
+    ollama_client: ollama.Client,
 ) -> int:
     """
-    Counts the number of tokens in a given text using a specified Ollama model
-    by calling the Ollama server's /api/tokenize endpoint.
+    Counts the number of tokens in a given text for a specified Ollama model
+    by making a minimal request to the /api/generate endpoint and extracting
+    the 'prompt_eval_count' from the response.
+
+    This method is generally more accurate for the specific Ollama model instance
+    than using an external tokenizer, but it incurs the overhead of an API call
+    and model processing for the prompt.
 
     Args:
-        text_to_tokenize (str): The text to be tokenized.
-        model_name (str): The name of the Ollama model to use (e.g., "llama3", "mistral").
-        ollama_host (str): The base URL of the Ollama server (default: "http://localhost:11434").
-        timeout (int): Timeout for the request in seconds (default: 30).
-        verify_ssl_certificate (bool): Whether to verify SSL.
-        headers (Optional[Dict[str, str]]): Optional headers for the request.
+        text_to_tokenize: The string to tokenize.
+        model_name: The name of the Ollama model (e.g., "llama3:8b", "mistral").
+        ollama_host: The URL of the Ollama API host.
+        timeout: Timeout for the request to Ollama.
+        verify_ssl_certificate: Whether to verify SSL certificates for the Ollama host.
+        headers: Optional custom headers for the request to Ollama.
+        num_predict_for_eval: How many tokens to ask the model to "predict" to get
+                              the prompt evaluation count. 0 is usually sufficient and most efficient.
+                              If 0 doesn't consistently yield `prompt_eval_count`, try 1.
 
     Returns:
-        int: The number of tokens. Returns -1 if an error occurs.
-    """
-    api_url = f"{ollama_host.rstrip('/')}/api/tokenize"
-    payload = {
-        "model": model_name,
-        "prompt": text_to_tokenize
-    }
-    request_headers = headers if headers else {}
-
-    try:
-        response = requests.post(api_url, json=payload, timeout=timeout, verify=verify_ssl_certificate, headers=request_headers)
-        response.raise_for_status()  # Raises HTTPError for bad responses (4xx or 5xx)
-
-        response_data = response.json()
-
-        if "tokens" in response_data and isinstance(response_data["tokens"], list):
-            return len(response_data["tokens"])
-        else:
-            ASCIIColors.warning(
-                f"Ollama response for token count did not contain a 'tokens' list. Response: {response_data}"
-            )
-            return -1 # Or raise ValueError
-
-    except requests.exceptions.HTTPError as http_err:
-        ASCIIColors.error(f"HTTP error occurred during token count: {http_err} - {http_err.response.text if http_err.response else 'No response text'}")
-        return -1
-    except requests.exceptions.RequestException as req_err:
-        ASCIIColors.error(f"Request error occurred during token count: {req_err}")
-        return -1
-    except json.JSONDecodeError as json_err:
-        ASCIIColors.error(
-            f"Failed to decode JSON response from Ollama during token count: {json_err}. Response text: {response.text if hasattr(response, 'text') else 'No response object'}"
-        )
-        return -1
-    except Exception as e:
-        ASCIIColors.error(f"An unexpected error occurred during token count: {e}")
-        return -1
+        The number of tokens as reported by 'prompt_eval_count'.
 
+    Raises:
+        requests.exceptions.RequestException: If the API request fails.
+        KeyError: If 'prompt_eval_count' is not found in the response.
+        json.JSONDecodeError: If the response is not valid JSON.
+        RuntimeError: For other operational errors.
+    """
+    res = ollama_client.chat(
+                        model=model_name,
+                        messages=[{"role":"system","content":""},{"role":"user", "content":text_to_tokenize}],
+                        stream=False,options={"num_predict":1}                        
+                    )
+    
+    return res.prompt_eval_count-5
 class OllamaBinding(LollmsLLMBinding):
     """Ollama-specific binding implementation using the ollama-python library."""
     
@@ -132,6 +115,7 @@ class OllamaBinding(LollmsLLMBinding):
                      images: Optional[List[str]] = None, # List of image file paths
                      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
@@ -191,7 +175,7 @@ class OllamaBinding(LollmsLLMBinding):
                     # If images were base64 strings, they would need decoding to bytes first.
                     processed_images.append(img_path)
 
-                messages = [{'role': 'user', 'content': prompt, 'images': processed_images if processed_images else None}]
+                messages = [{'role': 'system', 'content':system_prompt},{'role': 'user', 'content': prompt, 'images': processed_images if processed_images else None}]
                 
                 if stream:
                     response_stream = self.ollama_client.chat(
@@ -253,53 +237,31 @@ 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:
         """
@@ -314,8 +276,8 @@ 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.host_address, verify_ssl_certificate=self.verify_ssl_certificate, headers=self.ollama_client_headers)
-    
+        #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.
@@ -334,7 +296,7 @@ class OllamaBinding(LollmsLLMBinding):
         if not self.ollama_client:
              raise Exception("Ollama client not initialized.")
 
-        model_to_use = kwargs.get("model", self.model_name)
+        model_to_use = kwargs.get("model", "bge-m3")
         if not model_to_use:
             raise ValueError("Model name for embedding must be specified either in init or via kwargs.")
             
@@ -574,4 +536,4 @@ if __name__ == '__main__':
         ASCIIColors.error(f"An error occurred during testing: {e}")
         trace_exception(e)
 
-    ASCIIColors.yellow("\nOllamaBinding test finished.")
+    ASCIIColors.yellow("\nOllamaBinding test finished.")

lollms_client/lollms_core.py

@@ -278,7 +278,20 @@ class LollmsClient():
         if self.binding:
             return self.binding.detokenize(tokens)
         raise RuntimeError("LLM binding not initialized.")
+    def count_tokens(self, text: str) -> int:
+        """
+        Counts how many tokens are there in the text using the active LLM binding.
+
+        Args:
+            text (str): The text to tokenize.
 
+        Returns:
+            int: Number of tokens.
+        """
+        if self.binding:
+            return self.binding.count_tokens(text)
+        raise RuntimeError("LLM binding not initialized.")
+    
     def get_model_details(self) -> dict:
         """
         Get model information from the active LLM binding.
@@ -1611,4 +1624,4 @@ def chunk_text(text, tokenizer, detokenizer, chunk_size, overlap, use_separators
             if current_pos >= len(text):
                 break
 
-    return chunks
+    return chunks

{lollms_client-0.13.0.dist-info → lollms_client-0.13.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 0.13.0
+Version: 0.13.2
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License: Apache Software License
@@ -19,6 +19,13 @@ Requires-Python: >=3.7
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: requests
+Requires-Dist: ascii-colors
+Requires-Dist: pipmaster
+Requires-Dist: pyyaml
+Requires-Dist: tiktoken
+Requires-Dist: pydantic
+Requires-Dist: numpy
+Requires-Dist: pillow
 Dynamic: license-file
 
 # lollms_client

{lollms_client-0.13.0.dist-info → lollms_client-0.13.2.dist-info}/RECORD

@@ -1,4 +1,4 @@
-examples/simple_text_gen_test.py,sha256=CqBvkG7Zm2ya13AxgRKlhV1M4PiqeSOi9--yKL8MZ-E,8596
+examples/simple_text_gen_test.py,sha256=RoX9ZKJjGMujeep60wh5WT_GoBn0O9YKJY6WOy-ZmOc,8710
 examples/simple_text_gen_with_image_test.py,sha256=Euv53jbKTVJDvs854lgJvA5F-iRnAATLxAklig24ots,8534
 examples/text_2_audio.py,sha256=MfL4AH_NNwl6m0I0ywl4BXRZJ0b9Y_9fRqDIe6O-Sbw,3523
 examples/text_2_image.py,sha256=Ri7lQ-GW54YWQh2eofcaN6LpwFoorbpJsJffrcXl3cg,6415
@@ -12,9 +12,9 @@ examples/personality_test/chat_test.py,sha256=o2jlpoddFc-T592iqAiA29xk3x27KsdK5D
 examples/personality_test/chat_with_aristotle.py,sha256=4X_fwubMpd0Eq2rCReS2bgVlUoAqJprjkLXk2Jz6pXU,1774
 examples/personality_test/tesks_test.py,sha256=7LIiwrEbva9WWZOLi34fsmCBN__RZbPpxoUOKA_AtYk,1924
 examples/test_local_models/local_chat.py,sha256=slakja2zaHOEAUsn2tn_VmI4kLx6luLBrPqAeaNsix8,456
-lollms_client/__init__.py,sha256=5Rs-Uis0zGVg94icdyXKveRFsva8Vh1otWtGDKEfSsw,821
+lollms_client/__init__.py,sha256=y-N8Dw10pI9pHtP_zlVzsj7bVjsu873EdOlbqxaxZRU,823
 lollms_client/lollms_config.py,sha256=goEseDwDxYJf3WkYJ4IrLXwg3Tfw73CXV2Avg45M_hE,21876
-lollms_client/lollms_core.py,sha256=_iTQY64ePop-6YxA2Xb90CxuiF9gzz9c5SHcLUsJxVc,77559
+lollms_client/lollms_core.py,sha256=ZTbEVn1M_gHAL3mL5mf3wGYAXidAtnSI3qEjwz2HlwY,77980
 lollms_client/lollms_discussion.py,sha256=9b83m0D894jwpgssWYTQHbVxp1gJoI-J947Ui_dRXII,2073
 lollms_client/lollms_functions.py,sha256=p8SFtmEPqvVCsIz2fZ5HxyOHaxjrAo5c12uTzJnb6m8,3594
 lollms_client/lollms_js_analyzer.py,sha256=01zUvuO2F_lnUe_0NLxe1MF5aHE1hO8RZi48mNPv-aw,8361
@@ -30,7 +30,7 @@ lollms_client/lollms_types.py,sha256=cfc1sremM8KR4avkYX99fIVkkdRvXErrCWKGjLrgv50
 lollms_client/lollms_utilities.py,sha256=YAgamfp0pBVApR68AHKjhp1lh6isMNF8iadwWLl63c0,7045
 lollms_client/llm_bindings/__init__.py,sha256=9sWGpmWSSj6KQ8H4lKGCjpLYwhnVdL_2N7gXCphPqh4,14
 lollms_client/llm_bindings/lollms/__init__.py,sha256=l1q2KnMQALz9QpLa3OUQ8e29KU4RCwkrmrdBvd7Z_kc,12236
-lollms_client/llm_bindings/ollama/__init__.py,sha256=mKGLeoRVpKW1YW4fnLQ4KlgbgHsN4i5TTfG3B4CxwVA,28428
+lollms_client/llm_bindings/ollama/__init__.py,sha256=DyueED1cJmmJFg5evYmu-lrkwsN9pAxaVcwgUkcAZHU,26467
 lollms_client/llm_bindings/openai/__init__.py,sha256=SWBgnOcOWmFRSKTN1S9ATownHNBJ9f6FEtI3L4xNJNM,11861
 lollms_client/llm_bindings/tensor_rt/__init__.py,sha256=ZpeSKAbN8rh6zkysYl95sXG9Ci702NuPAhXC6zb1zT4,31840
 lollms_client/llm_bindings/transformers/__init__.py,sha256=8JbX3B-obLt5NNtcNOGD_E0f8OQTma2pNYtVt2urTOM,12572
@@ -45,8 +45,8 @@ lollms_client/tts_bindings/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJ
 lollms_client/tts_bindings/lollms/__init__.py,sha256=8x2_T9XscvISw2TiaLoFxvrS7TIsVLdqbwSc04cX-wc,7164
 lollms_client/ttv_bindings/__init__.py,sha256=UZ8o2izQOJLQgtZ1D1cXoNST7rzqW22rL2Vufc7ddRc,3141
 lollms_client/ttv_bindings/lollms/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lollms_client-0.13.0.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
-lollms_client-0.13.0.dist-info/METADATA,sha256=I5sQiDvtijyYe72vvQWBjecdglA_-oyfY4I9W23eUZw,7103
-lollms_client-0.13.0.dist-info/WHEEL,sha256=DnLRTWE75wApRYVsjgc6wsVswC54sMSJhAEd4xhDpBk,91
-lollms_client-0.13.0.dist-info/top_level.txt,sha256=NI_W8S4OYZvJjb0QWMZMSIpOrYzpqwPGYaklhyWKH2w,23
-lollms_client-0.13.0.dist-info/RECORD,,
+lollms_client-0.13.2.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+lollms_client-0.13.2.dist-info/METADATA,sha256=IsZiVKLRi7NaMhITn6pFz3CX5zpIJAvb8vsddYIBJQA,7276
+lollms_client-0.13.2.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
+lollms_client-0.13.2.dist-info/top_level.txt,sha256=NI_W8S4OYZvJjb0QWMZMSIpOrYzpqwPGYaklhyWKH2w,23
+lollms_client-0.13.2.dist-info/RECORD,,

{lollms_client-0.13.0.dist-info → lollms_client-0.13.2.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.4.0)
+Generator: setuptools (80.7.1)
 Root-Is-Purelib: true
 Tag: py3-none-any