models/llm_setup.py

import torch
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM

def setup_llm():
    """Set up a more capable LLM for CSV analysis."""
    try:
        # Try to load FLAN-T5-small, which is better for instruction following
        # while still being relatively small (~300MB)
        model_name = "google/flan-t5-small"
        
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
        
        generator = pipeline(
            "text2text-generation",
            model=model,
            tokenizer=tokenizer,
            max_length=512
        )
        
        # Create a wrapper class that matches the expected interface
        class FlanT5LLM:
            def complete(self, prompt):
                class Response:
                    def __init__(self, text):
                        self.text = text
                
                try:
                    # For FLAN-T5, we don't need to strip the prompt from the output
                    result = generator(prompt, max_length=150, do_sample=False)[0]
                    response_text = result["generated_text"].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)}")
        
        return FlanT5LLM()
    
    except Exception as e:
        print(f"Error setting up FLAN-T5 model: {e}")
        
        # Fallback to a simpler model if FLAN-T5 fails
        try:
            # Try T5-small as a fallback
            model_name = "t5-small"
            
            tokenizer = AutoTokenizer.from_pretrained(model_name)
            model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
            
            generator = pipeline(
                "text2text-generation",
                model=model,
                tokenizer=tokenizer,
                max_length=512
            )
            
            class T5LLM:
                def complete(self, prompt):
                    class Response:
                        def __init__(self, text):
                            self.text = text
                    
                    try:
                        result = generator(prompt, max_length=150, do_sample=False)[0]
                        return Response(result["generated_text"].strip())
                    except Exception as e:
                        return Response(f"Error: {str(e)}")
            
            return T5LLM()
            
        except Exception as e2:
            print(f"Error setting up fallback model: {e2}")
            
            # Last resort - dummy LLM
            class DummyLLM:
                def complete(self, prompt):
                    class Response:
                        def __init__(self, text):
                            self.text = text
                    
                    return Response("Model initialization failed. Please check logs.")
            
            return DummyLLM()