Update llm_handler.py

llm_handler.py

@@ -1,34 +1,31 @@
-from openai import OpenAI
-import llamanet
+from llama_cpp import Llama
+from llama_cpp_agent import LlamaCppAgent
+from llama_cpp_agent import MessagesFormatterType
+from llama_cpp_agent.providers import LlamaCppPythonProvider
 
-# Initialize LlamaNet
-llamanet.run()
+# Initialize the Llama model
+llama_model = Llama("Arcee-Spark-GGUF/Arcee-Spark-Q4_K_M.gguf", n_batch=1024, n_threads=10, n_gpu_layers=33, n_ctx=2048, verbose=False)
 
-# Create an instance of the OpenAI class (used by LlamaNet)
-client = OpenAI()
+# Create the provider
+provider = LlamaCppPythonProvider(llama_model)
 
-def send_to_llamanet(msg_list):
+# Create the agent
+agent = LlamaCppAgent(
+    provider,
+    system_prompt="You are a helpful assistant.",
+    predefined_messages_formatter_type=MessagesFormatterType.CHATML,
+    debug_output=True
+)
+
+# Set provider settings
+settings = provider.get_provider_default_settings()
+settings.max_tokens = 2000
+settings.stream = True
+
+def send_to_llm(provider, msg_list):
     try:
-        completion = client.chat.completions.create(
-            model="https://huggingface.co/arcee-ai/Arcee-Spark-GGUF/blob/main/Arcee-Spark-IQ4_XS.gguf",
-            messages=msg_list,
-            temperature=0.6,
-            stream=True
-        )
-        
-        response = ""
-        for chunk in completion:
-            if chunk.choices[0].delta.content is not None:
-                response += chunk.choices[0].delta.content
-        
-        return response, None  # LlamaNet doesn't provide usage info
-    
+        response = agent.chat(msg_list, settings=settings)
+        return response.content, None  # We don't have usage info in this case
     except Exception as e:
-        print(f"Error in send_to_llamanet: {str(e)}")
+        print(f"Error in send_to_llm: {str(e)}")
         return f"Error: {str(e)}", None
-
-def send_to_llm(provider, msg_list):
-    if provider == "llamanet":
-        return send_to_llamanet(msg_list)
-    else:
-        raise ValueError(f"Unknown provider: {provider}")