src/llm_generator.py

from typing import Optional, List, Dict, Generator
from groq import Groq

from config import (
    GROQ_API_KEY,
    LLM_MODELS,
    DEFAULT_LLM_MODEL,
)
from prompts import SYSTEM_PROMPT_TR


class LLMGenerator:
    
    def __init__(self, api_key: str = None):
        self._api_key = api_key or GROQ_API_KEY
        if not self._api_key:
            raise ValueError("API key required.")
        self._client = None
        self.available_models = LLM_MODELS
    
    @property
    def client(self) -> Groq:
        if self._client is None:
            self._client = Groq(api_key=self._api_key)
        return self._client
    
    def _format_chat_history(self, chat_history: List[Dict[str, str]]) -> str:
        if not chat_history:
            return "Henüz konuşma geçmişi yok."
        
        parts = []
        for msg in chat_history:
            role = "Kullanici" if msg["role"] == "user" else "Asistan"
            content = msg["content"][:500]
            parts.append(f"{role}: {content}")
        
        return "\n".join(parts)
    
    def _build_prompt(
        self,
        question: str,
        context: str,
        chat_history: Optional[List[Dict[str, str]]] = None
    ) -> str:
        history_str = self._format_chat_history(chat_history or [])
        
        return SYSTEM_PROMPT_TR.format(
            context=context,
            question=question,
            chat_history=history_str
        )
    
    def generate(
        self,
        question: str,
        context: str,
        chat_history: Optional[List[Dict[str, str]]] = None,
        model_id: str = DEFAULT_LLM_MODEL,
        temperature: float = 0.1,
        max_tokens: int = 1024
    ) -> str:
        prompt = self._build_prompt(question, context, chat_history)
        
        try:
            response = self.client.chat.completions.create(
                model=model_id,
                messages=[
                    {"role": "system", "content": "Sen Türkçe yanıt veren bir uzman asistansın."},
                    {"role": "user", "content": prompt}
                ],
                temperature=temperature,
                max_tokens=max_tokens
            )
            return response.choices[0].message.content
        except Exception as e:
            return f"Hata oluştu: {str(e)}"
    
    def generate_stream(
        self,
        question: str,
        context: str,
        chat_history: Optional[List[Dict[str, str]]] = None,
        model_id: str = DEFAULT_LLM_MODEL,
        temperature: float = 0.1,
        max_tokens: int = 1024
    ) -> Generator[str, None, None]:
        prompt = self._build_prompt(question, context, chat_history)
        
        try:
            stream = self.client.chat.completions.create(
                model=model_id,
                messages=[
                    {"role": "system", "content": "Sen Türkçe yanıt veren bir uzman asistansın."},
                    {"role": "user", "content": prompt}
                ],
                temperature=temperature,
                max_tokens=max_tokens,
                stream=True
            )
            
            for chunk in stream:
                if chunk.choices[0].delta.content:
                    yield chunk.choices[0].delta.content
        except Exception as e:
            yield f"Hata oluştu: {str(e)}"
    
    @staticmethod
    def get_model_display_name(model_id: str) -> str:
        for name, mid in LLM_MODELS.items():
            if mid == model_id:
                return name
        return model_id
    
    @staticmethod
    def get_model_id(display_name: str) -> str:
        return LLM_MODELS.get(display_name, DEFAULT_LLM_MODEL)


_llm_generator_instance = None


def get_llm_generator() -> LLMGenerator:
    global _llm_generator_instance
    if _llm_generator_instance is None:
        _llm_generator_instance = LLMGenerator()
    return _llm_generator_instance


def reset_llm_generator():
    global _llm_generator_instance
    _llm_generator_instance = None


if __name__ == "__main__":
    generator = LLMGenerator()
    
    context = "Atlas ERP sistemi muhasebe ve finans modülleri içerir."
    question = "Atlas nedir?"
    
    response = generator.generate(question, context)
    print(f"Response: {response}")