llm_handler.py

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 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 the provider
provider = LlamaCppPythonProvider(llama_model)

# 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:
        # Convert the msg_list to the correct format expected by the agent
        formatted_messages = [
            {"role": msg["role"], "content": msg["content"]}
            for msg in msg_list
        ]

        # Call get_chat_response with the formatted messages
        response = agent.get_chat_response(formatted_messages, llm_sampling_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_llm: {str(e)}")
        return f"Error: {str(e)}", None