model_api.py

from huggingface_hub import InferenceClient
import os
import requests
import json

LLAMA_MODEL = "meta-llama/Llama-3.2-3B-Instruct"  # Default model
USE_LOCAL_OLLAMA = False  # Set to True if using local Ollama

def query_model(prompt):
    """
    Query the Llama model with the given prompt
    Supports both Hugging Face Inference API and local Ollama
    """
    try:
        if USE_LOCAL_OLLAMA:
            return query_ollama(prompt)
        else:
            return query_huggingface(prompt)
        
    except Exception as e:
        return f"Error generating workout plan: {str(e)}"

def query_huggingface(prompt):
    """
    Query Llama via Hugging Face Inference API
    """
    HF_TOKEN = os.getenv("HF_TOKEN")
    
    if not HF_TOKEN:
        return "Error: HF_TOKEN not found. Please set your Hugging Face token in environment variables."
    
    # Initialize the client with Llama model
    client = InferenceClient(
        model=LLAMA_MODEL,
        token=HF_TOKEN
    )
    
    # Enhanced system prompt for better responses
    system_prompt = """You are a certified professional fitness trainer with expertise in creating personalized workout plans. 
    Always provide complete, detailed workout plans with:
    - Clear day-by-day structure
    - Specific exercises with sets, reps, and rest periods
    - Warm-up and cool-down recommendations
    - Safety considerations based on user's profile
    When asked for a 5-day plan, ensure ALL 5 days are included with clear day headers."""
    
    # Make the API call
    response = client.chat_completion(
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": prompt}
        ],
        max_tokens=3000,
        temperature=0.7,
        top_p=0.95
    )
    
    # Extract and return the response
    workout_plan = response.choices[0].message.content
    
    # Verify if the response contains all 5 days
    days_found = sum([f"Day {i}" in workout_plan for i in range(1, 6)])
    
    if days_found < 5:
        # If incomplete, try one more time with more explicit instruction
        retry_prompt = prompt + "\n\nIMPORTANT: The previous response was incomplete. Please ensure ALL 5 days (Day 1 through Day 5) are included in the plan. Each day should be clearly marked with 'Day X' header and include 4-6 exercises."
        
        retry_response = client.chat_completion(
            messages=[
                {"role": "system", "content": system_prompt},
                {"role": "user", "content": retry_prompt}
            ],
            max_tokens=3000,
            temperature=0.7
        )
        workout_plan = retry_response.choices[0].message.content
    
    return workout_plan

def query_ollama(prompt):
    """
    Query Llama via local Ollama (completely free, no API key needed)
    """
    try:
        response = requests.post(
            "http://localhost:11434/api/generate",
            json={
                "model": "llama3.2:3b",  # or "llama3.2:1b" for lighter model
                "prompt": f"""You are a certified professional fitness trainer. Create a comprehensive 5-day workout plan.

{prompt}

Provide a complete, detailed 5-day workout plan with clear day headers, exercises, sets, reps, and rest periods.""",
                "stream": False,
                "max_tokens": 3000,
                "temperature": 0.7
            }
        )
        
        if response.status_code == 200:
            return response.json()["response"]
        else:
            return f"Error: Ollama returned status code {response.status_code}"
            
    except requests.exceptions.ConnectionError:
        return "Error: Cannot connect to Ollama. Make sure Ollama is running locally (run 'ollama serve' in terminal)"
    except Exception as e:
        return f"Error with Ollama: {str(e)}"

def test_api_connection():
    """
    Test function to verify API connection
    """
    try:
        if USE_LOCAL_OLLAMA:
            # Test Ollama connection
            response = requests.post(
                "http://localhost:11434/api/generate",
                json={
                    "model": "llama3.2:3b",
                    "prompt": "Say 'API connection successful' if you can read this.",
                    "stream": False,
                    "max_tokens": 50
                }
            )
            if response.status_code == 200:
                return True, "Ollama connection successful"
            else:
                return False, f"Ollama connection failed: {response.status_code}"
        else:
            # Test Hugging Face connection
            HF_TOKEN = os.getenv("HF_TOKEN")
            if not HF_TOKEN:
                return False, "HF_TOKEN not found"
            
            client = InferenceClient(
                model=LLAMA_MODEL,
                token=HF_TOKEN
            )
            
            response = client.chat_completion(
                messages=[
                    {"role": "system", "content": "You are a helpful assistant."},
                    {"role": "user", "content": "Say 'API connection successful' if you can read this."}
                ],
                max_tokens=50,
                temperature=0.1
            )
            
            return True, f"API connection successful (using {LLAMA_MODEL})"
        
    except Exception as e:
        return False, f"API connection failed: {str(e)}"

def switch_model(model_name):
    """
    Switch to a different Llama model
    """
    global LLAMA_MODEL
    LLAMA_MODEL = model_name
    return f"Switched to {model_name}"

def set_ollama_mode(use_ollama):
    """
    Switch between Hugging Face API and local Ollama
    """
    global USE_LOCAL_OLLAMA
    USE_LOCAL_OLLAMA = use_ollama
    mode = "local Ollama" if use_ollama else "Hugging Face API"
    return f"Switched to {mode} mode"