lollms-client 0.32.0__py3-none-any.whl → 0.32.1__py3-none-any.whl

lollms_client/__init__.py

@@ -8,7 +8,7 @@ from lollms_client.lollms_utilities import PromptReshaper # Keep general utiliti
 from lollms_client.lollms_mcp_binding import LollmsMCPBinding, LollmsMCPBindingManager
 from lollms_client.lollms_llm_binding import LollmsLLMBindingManager
 
-__version__ = "0.32.0" # Updated version
+__version__ = "0.32.1" # Updated version
 
 # Optionally, you could define __all__ if you want to be explicit about exports
 __all__ = [

lollms_client/llm_bindings/llamacpp/__init__.py

@@ -901,6 +901,77 @@ class LlamaCppServerBinding(LollmsLLMBinding):
     def __del__(self):
         self.unload_model()
 
+    def get_ctx_size(self, model_name: Optional[str] = None) -> Optional[int]:
+        """
+        Retrieves context size for a model from a hardcoded list.
+
+        This method checks if the model name contains a known base model identifier
+        (e.g., 'llama3.1', 'gemma2') to determine its context length. It's intended
+        as a failsafe when the context size cannot be retrieved directly from the
+        Ollama API.
+        """
+        if model_name is None:
+            model_name = self.model_name
+
+        # Hardcoded context sizes for popular models. More specific names (e.g., 'llama3.1')
+        # should appear, as they will be checked first due to the sorting logic below.
+        known_contexts = {
+            'llama3.1': 131072,   # Llama 3.1 extended context
+            'llama3.2': 131072,   # Llama 3.2 extended context
+            'llama3.3': 131072,   # Assuming similar to 3.1/3.2
+            'llama3': 8192,       # Llama 3 default
+            'llama2': 4096,       # Llama 2 default
+            'mixtral8x22b': 65536, # Mixtral 8x22B default
+            'mixtral': 32768,     # Mixtral 8x7B default
+            'mistral': 32768,     # Mistral 7B v0.2+ default
+            'gemma3': 131072,     # Gemma 3 with 128K context
+            'gemma2': 8192,       # Gemma 2 default
+            'gemma': 8192,        # Gemma default
+            'phi3': 131072,       # Phi-3 variants often use 128K (mini/medium extended)
+            'phi2': 2048,         # Phi-2 default
+            'phi': 2048,          # Phi default (older)
+            'qwen2.5': 131072,    # Qwen2.5 with 128K
+            'qwen2': 32768,       # Qwen2 default for 7B
+            'qwen': 8192,         # Qwen default
+            'codellama': 16384,   # CodeLlama extended
+            'codegemma': 8192,    # CodeGemma default
+            'deepseek-coder-v2': 131072,  # DeepSeek-Coder V2 with 128K
+            'deepseek-coder': 16384,  # DeepSeek-Coder V1 default
+            'deepseek-v2': 131072,    # DeepSeek-V2 with 128K
+            'deepseek-llm': 4096,     # DeepSeek-LLM default
+            'yi1.5': 32768,       # Yi-1.5 with 32K
+            'yi': 4096,           # Yi base default
+            'command-r': 131072,  # Command-R with 128K
+            'wizardlm2': 32768,   # WizardLM2 (Mistral-based)
+            'wizardlm': 16384,    # WizardLM default
+            'zephyr': 65536,      # Zephyr beta (Mistral-based extended)
+            'vicuna': 2048,       # Vicuna default (up to 16K in some variants)
+            'falcon': 2048,       # Falcon default
+            'starcoder': 8192,    # StarCoder default
+            'stablelm': 4096,     # StableLM default
+            'orca2': 4096,        # Orca 2 default
+            'orca': 4096,         # Orca default
+            'dolphin': 32768,     # Dolphin (often Mistral-based)
+            'openhermes': 8192,   # OpenHermes default
+        }
+
+        normalized_model_name = model_name.lower().strip()
+
+        # Sort keys by length in descending order. This ensures that a more specific
+        # name like 'llama3.1' is checked before a less specific name like 'llama3'.
+        sorted_base_models = sorted(known_contexts.keys(), key=len, reverse=True)
+
+        for base_name in sorted_base_models:
+            if base_name in normalized_model_name:
+                context_size = known_contexts[base_name]
+                ASCIIColors.warning(
+                    f"Using hardcoded context size for model '{model_name}' "
+                    f"based on base name '{base_name}': {context_size}"
+                )
+                return context_size
+
+        ASCIIColors.warning(f"Context size not found for model '{model_name}' in the hardcoded list.")
+        return None
 
 if __name__ == '__main__':
     global full_streamed_text # Define for the callback

lollms_client/llm_bindings/ollama/__init__.py

@@ -598,6 +598,94 @@ class OllamaBinding(LollmsLLMBinding):
         ASCIIColors.info(f"Ollama model set to: {model_name}. It will be loaded by the server on first use.")
         return True
 
+    def get_ctx_size(self, model_name: Optional[str] = None) -> Optional[int]:
+        """
+        Retrieves the context size for an Ollama model.
+        
+        The effective context size is the `num_ctx` parameter if overridden in the Modelfile,
+        otherwise it falls back to the model's default context length from its architecture details.
+        As a final failsafe, uses a hardcoded list of known popular models' context lengths.
+        """
+        if model_name is None:
+            model_name = self.model_name
+            
+        try:
+            info = ollama.show(model_name)
+            
+            # Parse num_ctx from the 'parameters' string (e.g., "PARAMETER num_ctx 4096")
+            parameters = info.get('parameters', '')
+            num_ctx = None
+            for param in parameters.split('\n'):
+                if param.strip().startswith('num_ctx'):
+                    num_ctx = int(param.split()[1])
+                    break
+            
+            if num_ctx is not None:
+                return num_ctx
+            
+            # Fall back to model_info context_length (e.g., 'llama.context_length')
+            model_info = info.get('model_info', {})
+            arch = model_info.get('general.architecture', '')
+            context_key = f'{arch}.context_length' if arch else 'general.context_length'
+            context_length = model_info.get(context_key)
+            
+            if context_length is not None:
+                return int(context_length)
+            
+        except Exception as e:
+            ASCIIColors.warning(f"Error fetching model info: {str(e)}")
+        
+        # Failsafe: Hardcoded context sizes for popular Ollama models
+        known_contexts = {
+            'llama2': 4096,       # Llama 2 default
+            'llama3': 8192,       # Llama 3 default
+            'llama3.1': 131072,   # Llama 3.1 extended context
+            'llama3.2': 131072,   # Llama 3.2 extended context
+            'llama3.3': 131072,   # Assuming similar to 3.1/3.2
+            'mistral': 32768,     # Mistral 7B v0.2+ default
+            'mixtral': 32768,     # Mixtral 8x7B default
+            'mixtral8x22b': 65536, # Mixtral 8x22B default
+            'gemma': 8192,        # Gemma default
+            'gemma2': 8192,       # Gemma 2 default
+            'gemma3': 131072,     # Gemma 3 with 128K context
+            'phi': 2048,          # Phi default (older)
+            'phi2': 2048,         # Phi-2 default
+            'phi3': 131072,       # Phi-3 variants often use 128K (mini/medium extended)
+            'qwen': 8192,         # Qwen default
+            'qwen2': 32768,       # Qwen2 default for 7B
+            'qwen2.5': 131072,    # Qwen2.5 with 128K
+            'codellama': 16384,   # CodeLlama extended
+            'codegemma': 8192,    # CodeGemma default
+            'deepseek-coder': 16384,  # DeepSeek-Coder V1 default
+            'deepseek-coder-v2': 131072,  # DeepSeek-Coder V2 with 128K
+            'deepseek-llm': 4096,     # DeepSeek-LLM default
+            'deepseek-v2': 131072,    # DeepSeek-V2 with 128K
+            'yi': 4096,           # Yi base default
+            'yi1.5': 32768,       # Yi-1.5 with 32K
+            'command-r': 131072,  # Command-R with 128K
+            'vicuna': 2048,       # Vicuna default (up to 16K in some variants)
+            'wizardlm': 16384,    # WizardLM default
+            'wizardlm2': 32768,   # WizardLM2 (Mistral-based)
+            'zephyr': 65536,      # Zephyr beta (Mistral-based extended)
+            'falcon': 2048,       # Falcon default
+            'starcoder': 8192,    # StarCoder default
+            'stablelm': 4096,     # StableLM default
+            'orca': 4096,         # Orca default
+            'orca2': 4096,        # Orca 2 default
+            'dolphin': 32768,     # Dolphin (often Mistral-based)
+            'openhermes': 8192,   # OpenHermes default
+        }
+        
+        # Extract base model name (e.g., 'llama3' from 'llama3:8b-instruct')
+        base_name = model_name.split(':')[0].lower().strip()
+        
+        if base_name in known_contexts:
+            ASCIIColors.warning(f"Using hardcoded context size for model '{model_name}': {known_contexts[base_name]}")
+            return known_contexts[base_name]
+        
+        ASCIIColors.warning(f"Context size not found for model '{model_name}'")
+        return None
+        
 if __name__ == '__main__':
     global full_streamed_text
     # Example Usage (requires an Ollama server running)

{lollms_client-0.32.0.dist-info → lollms_client-0.32.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 0.32.0
+Version: 0.32.1
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License: Apache Software License

{lollms_client-0.32.0.dist-info → lollms_client-0.32.1.dist-info}/RECORD

@@ -29,7 +29,7 @@ examples/mcp_examples/openai_mcp.py,sha256=7IEnPGPXZgYZyiES_VaUbQ6viQjenpcUxGiHE
 examples/mcp_examples/run_remote_mcp_example_v2.py,sha256=bbNn93NO_lKcFzfIsdvJJijGx2ePFTYfknofqZxMuRM,14626
 examples/mcp_examples/run_standard_mcp_example.py,sha256=GSZpaACPf3mDPsjA8esBQVUsIi7owI39ca5avsmvCxA,9419
 examples/test_local_models/local_chat.py,sha256=slakja2zaHOEAUsn2tn_VmI4kLx6luLBrPqAeaNsix8,456
-lollms_client/__init__.py,sha256=7Vw58C6Ala6ESJmnETjGFRBG1C1H-3SKmTnfbL46eVI,1147
+lollms_client/__init__.py,sha256=5paEgQrICKx3_3wtk257FCKOwolkgmfFPJPwc9aYkWY,1147
 lollms_client/lollms_config.py,sha256=goEseDwDxYJf3WkYJ4IrLXwg3Tfw73CXV2Avg45M_hE,21876
 lollms_client/lollms_core.py,sha256=gDhpB62AluEmbVFvPm7vdnZgP2hGBymDLun57K1jrOM,177352
 lollms_client/lollms_discussion.py,sha256=TT-AIKMdEuNNBjrWgMLcww8z6vIETO0J3cnoQgb9dhU,85448
@@ -54,11 +54,11 @@ lollms_client/llm_bindings/grok/__init__.py,sha256=5tIf3348RgAEaSp6FdG-LM9N8R7aR
 lollms_client/llm_bindings/groq/__init__.py,sha256=zyWKM78qHwSt5g0Bb8Njj7Jy8CYuLMyplx2maOKFFpg,12218
 lollms_client/llm_bindings/hugging_face_inference_api/__init__.py,sha256=PxgeRqT8dpa9GZoXwtSncy9AUgAN2cDKrvp_nbaWq0E,14027
 lollms_client/llm_bindings/litellm/__init__.py,sha256=pNkwyRPeENvTM4CDh6Pj3kQfxHfhX2pvXhGJDjKjp30,12340
-lollms_client/llm_bindings/llamacpp/__init__.py,sha256=gER3lYd4Ez_Ng_yLO5zSZH52KgUosHXfdTr5W2wX_Jk,69000
+lollms_client/llm_bindings/llamacpp/__init__.py,sha256=uNqOoxFYnsgrYb-lVXQ0QrENWTJC5Np5NMTXfOAYoko,72800
 lollms_client/llm_bindings/lollms/__init__.py,sha256=scGHEKzlGX5fw2XwefVicsf28GrwgN3wU5nl4EPJ_Sk,24424
 lollms_client/llm_bindings/lollms_webui/__init__.py,sha256=Thoq3PJR2e03Y2Kd_FBb-DULJK0zT5-2ID1YIJLcPlw,17864
 lollms_client/llm_bindings/mistral/__init__.py,sha256=624Gr462yBh52ttHFOapKgJOn8zZ1vZcTEcC3i4FYt8,12750
-lollms_client/llm_bindings/ollama/__init__.py,sha256=_plx8cO3Bl9igmIEvTkJ6tkZ2imHS_L76hCHdJAGIhQ,36851
+lollms_client/llm_bindings/ollama/__init__.py,sha256=dXKHIeQCS9pz5AS07GF1eWj3ieWiz3aFOtxOX7yojbs,41314
 lollms_client/llm_bindings/open_router/__init__.py,sha256=v91BpNcuQCbbA6r82gbgMP8UYhSrJUMOf4UtOzEo18Q,13235
 lollms_client/llm_bindings/openai/__init__.py,sha256=Z0zNTfBgBGwkwArN375kBt4otrUTI_84pHgVuyuDy0c,26253
 lollms_client/llm_bindings/openllm/__init__.py,sha256=xv2XDhJNCYe6NPnWBboDs24AQ1VJBOzsTuMcmuQ6xYY,29864
@@ -93,9 +93,9 @@ lollms_client/tts_bindings/piper_tts/__init__.py,sha256=0IEWG4zH3_sOkSb9WbZzkeV5
 lollms_client/tts_bindings/xtts/__init__.py,sha256=FgcdUH06X6ZR806WQe5ixaYx0QoxtAcOgYo87a2qxYc,18266
 lollms_client/ttv_bindings/__init__.py,sha256=UZ8o2izQOJLQgtZ1D1cXoNST7rzqW22rL2Vufc7ddRc,3141
 lollms_client/ttv_bindings/lollms/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lollms_client-0.32.0.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+lollms_client-0.32.1.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
 test/test_lollms_discussion.py,sha256=KxTsV1bPdNz8QqZd7tIof9kTWkeXLUtAMU08BQmoY6U,16829
-lollms_client-0.32.0.dist-info/METADATA,sha256=uuOa7NYNuAQ8ibP6D24yqMQqg6Asl1Cep1UIIH1b6Dw,38717
-lollms_client-0.32.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lollms_client-0.32.0.dist-info/top_level.txt,sha256=1jIpjTnOSGEGtIW2rGAFM6tVRzgsDdMOiox_SmDH_zw,28
-lollms_client-0.32.0.dist-info/RECORD,,
+lollms_client-0.32.1.dist-info/METADATA,sha256=mw11ol3fFJpA2_sxa4olRXLkfVDeMfR_WCd63aLZdXw,38717
+lollms_client-0.32.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lollms_client-0.32.1.dist-info/top_level.txt,sha256=1jIpjTnOSGEGtIW2rGAFM6tVRzgsDdMOiox_SmDH_zw,28
+lollms_client-0.32.1.dist-info/RECORD,,