prompt_refiner.py

import json
import re
from typing import Optional, Dict, Any, Tuple
from pydantic import BaseModel, Field, validator
from huggingface_hub import InferenceClient
from huggingface_hub.errors import HfHubHTTPError
from variables import meta_prompts, prompt_refiner_model

class LLMResponse(BaseModel):
    initial_prompt_evaluation: str = Field(..., description="Evaluation of the initial prompt")
    refined_prompt: str = Field(..., description="The refined version of the prompt")
    explanation_of_refinements: str = Field(..., description="Explanation of the refinements made")
    response_content: Optional[Dict[str, Any]] = Field(None, description="Raw response content")

    @validator('initial_prompt_evaluation', 'refined_prompt', 'explanation_of_refinements')
    def clean_text_fields(cls, v):
        if isinstance(v, str):
            return v.strip().replace('\\n', '\n').replace('\\"', '"')
        return v

class PromptRefiner:
    def __init__(self, api_token: str, meta_prompts: dict):
        self.client = InferenceClient(token=api_token, timeout=120)
        self.meta_prompts = meta_prompts

    def refine_prompt(self, prompt: str, meta_prompt_choice: str) -> Tuple[str, str, str, dict]:
        """Refine the given prompt using the selected meta prompt."""
        try:
            selected_meta_prompt = self.meta_prompts.get(
                meta_prompt_choice, 
                self.meta_prompts["star"]
            )
            
            messages = [
                {
                    "role": "system", 
                    "content": 'You are an expert at refining and extending prompts. Given a basic prompt, provide a more relevant and detailed prompt.'
                },
                {
                    "role": "user", 
                    "content": selected_meta_prompt.replace("[Insert initial prompt here]", prompt)
                }
            ]
            
            response = self.client.chat_completion(
                model=prompt_refiner_model,
                messages=messages,
                max_tokens=3000,
                temperature=0.8
            )
            
            response_content = response.choices[0].message.content.strip()
            result = self._parse_response(response_content)
            
            try:
                llm_response = LLMResponse(**result)
                return (
                    llm_response.initial_prompt_evaluation,
                    llm_response.refined_prompt,
                    llm_response.explanation_of_refinements,
                    llm_response.dict()
                )
            except Exception as e:
                print(f"Error creating LLMResponse: {e}")
                return self._create_error_response(f"Error validating response: {str(e)}")

        except HfHubHTTPError as e:
            return self._create_error_response("Model timeout. Please try again later.")
        except Exception as e:
            return self._create_error_response(f"Unexpected error: {str(e)}")

    def _parse_response(self, response_content: str) -> dict:
        """Parse the LLM response content."""
        try:
            # Try to extract JSON from <json> tags
            json_match = re.search(r'<json>\s*(.*?)\s*</json>', response_content, re.DOTALL)
            if json_match:
                json_str = json_match.group(1).strip()
                # Clean up the JSON string
                json_str = re.sub(r'\s+', ' ', json_str)
                json_str = json_str.replace('•', '*')  # Replace bullet points
                
                try:
                    parsed_json = json.loads(json_str)
                    if isinstance(parsed_json, str):
                        parsed_json = json.loads(parsed_json)
                    
                    return {
                        "initial_prompt_evaluation": parsed_json.get("initial_prompt_evaluation", ""),
                        "refined_prompt": parsed_json.get("refined_prompt", ""),
                        "explanation_of_refinements": parsed_json.get("explanation_of_refinements", ""),
                        "response_content": parsed_json
                    }
                except json.JSONDecodeError as e:
                    print(f"JSON parsing error: {e}")
                    return self._create_error_dict(str(e))
            
            # Fallback to regex parsing if JSON extraction fails
            return self._parse_with_regex(response_content)

        except Exception as e:
            print(f"Error parsing response: {e}")
            print(f"Raw content: {response_content}")
            return self._create_error_dict(str(e))

    def _parse_with_regex(self, content: str) -> dict:
        """Parse content using regex patterns when JSON parsing fails."""
        output = {}
        for key in ["initial_prompt_evaluation", "refined_prompt", "explanation_of_refinements"]:
            pattern = rf'"{key}":\s*"(.*?)"(?:,|\}})'
            match = re.search(pattern, content, re.DOTALL)
            output[key] = match.group(1) if match else ""
        
        output["response_content"] = content
        return output

    def _create_error_dict(self, error_message: str) -> dict:
        """Create a standardized error response dictionary."""
        return {
            "initial_prompt_evaluation": f"Error parsing response: {error_message}",
            "refined_prompt": "",
            "explanation_of_refinements": "",
            "response_content": {"error": error_message}
        }

    def _create_error_response(self, error_message: str) -> Tuple[str, str, str, dict]:
        """Create a standardized error response tuple."""
        return (
            f"Error: {error_message}",
            "The selected model is currently unavailable.",
            "An error occurred during processing.",
            {"error": error_message}
        )