agent_core.py

"""
Agent Core Logic - Hugging Face Free Models
Uses free Inference API - no API keys needed!
"""

from huggingface_hub import InferenceClient


class HuggingFaceLLM:
    """
    LLM using free Hugging Face Inference API
    No API key needed for public models!
    """
    
    def __init__(self, model_name: str = None):
        # Free models available on HF
        self.available_models = [
            "meta-llama/Meta-Llama-3-8B-Instruct",
            "mistralai/Mistral-7B-Instruct-v0.2",
            "HuggingFaceH4/zephyr-7b-beta",
            "microsoft/Phi-3-mini-4k-instruct",
        ]
        
        self.model = model_name or self.available_models[0]
        self.client = InferenceClient()
        self.call_count = 0
        self.total_time = 0
        print(f"✅ LLM initialized: {self.model}")
    
    def call(self, prompt: str, max_tokens: int = 2000, temperature: float = 0.7) -> str:
        """Call Hugging Face Inference API"""
        import time
        
        try:
            self.call_count += 1
            start_time = time.time()
            
            # Try the primary model
            try:
                response = self.client.text_generation(
                    prompt=prompt,
                    model=self.model,
                    max_new_tokens=max_tokens,
                    temperature=temperature,
                    return_full_text=False
                )
                
                elapsed = time.time() - start_time
                self.total_time += elapsed
                
                return response
            
            except Exception as e:
                print(f"Primary model failed, trying fallback...")
                
                for fallback_model in self.available_models[1:]:
                    try:
                        response = self.client.text_generation(
                            prompt=prompt,
                            model=fallback_model,
                            max_new_tokens=max_tokens,
                            temperature=temperature,
                            return_full_text=False
                        )
                        self.model = fallback_model
                        
                        elapsed = time.time() - start_time
                        self.total_time += elapsed
                        
                        return response
                    except:
                        continue
                
                raise Exception("All models unavailable. Please try again later.")
                
        except Exception as e:
            return f"❌ Error: {str(e)}\n\nPlease try again in a moment."
    
    def get_stats(self) -> dict:
        """Get LLM usage statistics"""
        avg_time = self.total_time / self.call_count if self.call_count > 0 else 0
        return {
            "total_calls": self.call_count,
            "total_time": round(self.total_time, 2),
            "avg_time": round(avg_time, 2),
            "current_model": self.model
        }


class ImageForgeryAgent:
    """Main agent for Image Forgery Detection"""
    
    def __init__(self, llm_instance):
        self.llm = llm_instance
        self.config = {
            "user_name": "Denis",
            "competition": "Recod.ai/LUC Scientific Image Forgery Detection",
            "current_model": "EfficientNet-B4 UNet++",
            "current_score": 0.303,
            "target_score": 0.350,
            "version": "1.0.0"
        }
        self.query_count = 0
        self.successful_queries = 0
    
    def run(self, query: str) -> str:
        """Process query and return response"""
        self.query_count += 1
        
        try:
            system_context = f"""You are an expert AI assistant helping {self.config['user_name']} with the {self.config['competition']}.

Current Setup:
- Model: {self.config['current_model']}
- Score: {self.config['current_score']}
- Target: {self.config['target_score']}

Your Tools & Expertise:
1. suggest_improvements - Provide specific model optimization tips
2. compare_architectures - Compare different neural network architectures
3. debug_rle_masks - Help debug RLE encoding issues
4. create_strategy_plan - Create actionable competition strategies
5. analyze_discussions - Provide insights on competition trends

Instructions:
- Be specific and actionable
- Focus on practical Kaggle strategies
- Provide code examples when relevant
- Be concise but comprehensive
- Prioritize high-impact suggestions

User Question: {query}

Your Response:"""

            response = self.llm.call(system_context, max_tokens=1500, temperature=0.7)
            
            # Clean up response
            response = response.strip()
            
            # If response is too short, add context
            if len(response) < 50:
                response = "I'm processing your query. " + response
            
            self.successful_queries += 1
            return response
            
        except Exception as e:
            return f"❌ Error processing query: {str(e)}"
    
    def get_stats(self) -> dict:
        """Get agent statistics"""
        success_rate = (self.successful_queries / self.query_count * 100) if self.query_count > 0 else 0
        
        return {
            "total_queries": self.query_count,
            "successful_queries": self.successful_queries,
            "success_rate": round(success_rate, 1),
            "llm_stats": self.llm.get_stats()
        }


def get_llm():
    """Get LLM instance - no API key needed!"""
    return HuggingFaceLLM()