models/llm_setup_bk2.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

class SimpleTransformersLLM:
    """A simple wrapper for Hugging Face Transformers models."""
    
    def __init__(self, model_name="google/flan-t5-small"):
        """Initialize with a small model that works on CPU."""
        try:
            self.tokenizer = AutoTokenizer.from_pretrained(model_name)
            self.pipe = pipeline(
                "text-generation",
                model=model_name,
                tokenizer=self.tokenizer,
                max_length=512,
                device_map="auto"
            )
        except Exception as e:
            print(f"Error initializing model: {e}")
            self.pipe = None
    
    def complete(self, prompt):
        """Complete a prompt with the model."""
        class Response:
            def __init__(self, text):
                self.text = text
        
        if self.pipe is None:
            return Response("Model initialization failed.")
        
        try:
            result = self.pipe(prompt, max_length=len(prompt) + 200, do_sample=True)
            generated_text = result[0]["generated_text"]
            
            # Extract only the new text (not including the prompt)
            response_text = generated_text[len(prompt):].strip()
            if not response_text:
                response_text = "I couldn't generate a proper response."
                
            return Response(response_text)
        except Exception as e:
            print(f"Error generating response: {e}")
            return Response(f"Error generating response: {str(e)}")

def setup_llm():
    """Set up a simple LLM that doesn't require API keys."""
    try:
        # Try with a very small model first
        return SimpleTransformersLLM("google/flan-t5-small")
    except Exception as e:
        print(f"Error setting up LLM: {e}")
        
        # Fallback to dummy LLM
        class DummyLLM:
            def complete(self, prompt):
                class Response:
                    def __init__(self, text):
                        self.text = text
                
                return Response("This is a dummy response. The actual model couldn't be loaded.")
        
        return DummyLLM()