app.py

#!/usr/bin/env python3
"""
NEED AI - Production Flask API (FIXED VERSION)
Models are checked at runtime, not at import time
"""

from flask import Flask, request, jsonify
from flask_cors import CORS
import torch
import logging
import os
import time

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

app = Flask(__name__)
CORS(app)

# Set username as environment variable
os.environ['HF_USERNAME'] = 'yogami9'
HF_USERNAME = 'yogami9'
DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

logger.info(f"🖥️  Device: {DEVICE}")
logger.info(f"👤 HF Username: {HF_USERNAME}")

class ModelCache:
    """Lazy loading model cache"""
    
    def __init__(self):
        self.models = {}
        self.tokenizers = {}
        self.load_attempts = {}
        logger.info("📦 Model cache initialized")
    
    def _load_model(self, model_key, model_id, loader_func):
        """Generic model loader with error handling"""
        if model_key in self.models:
            return self.models[model_key], self.tokenizers.get(model_key)
        
        # Track load attempts to avoid infinite retries
        if self.load_attempts.get(model_key, 0) > 3:
            raise Exception(f"Max load attempts exceeded for {model_key}")
        
        self.load_attempts[model_key] = self.load_attempts.get(model_key, 0) + 1
        
        try:
            logger.info(f"📥 Loading {model_key} from {model_id}...")
            result = loader_func(model_id)
            logger.info(f"✅ {model_key} loaded successfully")
            return result
        except Exception as e:
            logger.error(f"❌ Failed to load {model_key}: {str(e)[:200]}")
            raise
    
    def get_category_model(self):
        def loader(model_id):
            from transformers import T5ForConditionalGeneration, T5Tokenizer
            model = T5ForConditionalGeneration.from_pretrained(model_id).to(DEVICE)
            tokenizer = T5Tokenizer.from_pretrained(model_id)
            self.models['category'] = model
            self.tokenizers['category'] = tokenizer
            return model, tokenizer
        
        return self._load_model('category', f'{HF_USERNAME}/need-category-recommendation', loader)
    
    def get_chat_model(self):
        def loader(model_id):
            from transformers import T5ForConditionalGeneration, T5Tokenizer
            model = T5ForConditionalGeneration.from_pretrained(model_id).to(DEVICE)
            tokenizer = T5Tokenizer.from_pretrained(model_id)
            self.models['chat'] = model
            self.tokenizers['chat'] = tokenizer
            return model, tokenizer
        
        return self._load_model('chat', f'{HF_USERNAME}/need-chat-support', loader)
    
    def get_service_model(self):
        def loader(model_id):
            from transformers import T5ForConditionalGeneration, T5Tokenizer
            model = T5ForConditionalGeneration.from_pretrained(model_id).to(DEVICE)
            tokenizer = T5Tokenizer.from_pretrained(model_id)
            self.models['service'] = model
            self.tokenizers['service'] = tokenizer
            return model, tokenizer
        
        return self._load_model('service', f'{HF_USERNAME}/need-service-description', loader)
    
    def get_search_model(self):
        def loader(model_id):
            from sentence_transformers import SentenceTransformer
            model = SentenceTransformer(model_id)
            self.models['search'] = model
            return model, None
        
        return self._load_model('search', f'{HF_USERNAME}/need-semantic-search', loader)
    
    def get_moderation_model(self):
        def loader(model_id):
            from transformers import AutoModelForSequenceClassification, AutoTokenizer
            model = AutoModelForSequenceClassification.from_pretrained(model_id).to(DEVICE)
            tokenizer = AutoTokenizer.from_pretrained(model_id)
            self.models['moderation'] = model
            self.tokenizers['moderation'] = tokenizer
            return model, tokenizer
        
        return self._load_model('moderation', f'{HF_USERNAME}/need-content-moderation', loader)

# Initialize cache
model_cache = ModelCache()

@app.route('/', methods=['GET'])
def home():
    return jsonify({
        'name': 'NEED AI API',
        'version': '2.0.1',
        'status': 'running',
        'username': HF_USERNAME,
        'models_loaded': len(model_cache.models),
        'endpoints': {
            'health': '/health',
            'category': '/api/category',
            'chat': '/api/chat',
            'service': '/api/service',
            'search': '/api/search',
            'moderate': '/api/moderate'
        },
        'documentation': 'https://github.com/Need-Service-App/need-ai-model',
        'note': 'First request per model takes 30-60 seconds (model download)'
    })

@app.route('/health', methods=['GET'])
def health():
    return jsonify({
        'status': 'healthy',
        'device': str(DEVICE),
        'gpu_available': torch.cuda.is_available(),
        'models_loaded': len(model_cache.models),
        'models_cached': list(model_cache.models.keys()),
        'username': HF_USERNAME,
        'note': 'Models load on first use'
    })

@app.route('/api/category', methods=['POST'])
def predict_category():
    try:
        start = time.time()
        data = request.get_json()
        
        if not data or 'query' not in data:
            return jsonify({'error': 'Missing "query" in request body'}), 400
        
        query = data['query']
        logger.info(f"📥 Category request: {query[:50]}")
        
        # Load model on demand
        model, tokenizer = model_cache.get_category_model()
        
        input_text = f"categorize: {query}"
        input_ids = tokenizer.encode(input_text, return_tensors="pt").to(DEVICE)
        
        with torch.no_grad():
            outputs = model.generate(input_ids, max_length=32, num_beams=4)
        
        category = tokenizer.decode(outputs[0], skip_special_tokens=True)
        elapsed = time.time() - start
        
        logger.info(f"✅ Category: {category} ({elapsed:.2f}s)")
        
        return jsonify({
            'query': query,
            'category': category,
            'inference_time': f'{elapsed:.3f}s',
            'model': 'category-recommendation'
        })
        
    except Exception as e:
        logger.error(f"❌ Category error: {str(e)}")
        return jsonify({
            'error': 'Inference failed',
            'message': str(e)[:300],
            'suggestion': 'Check logs. Model may be loading (wait 30s) or not accessible.'
        }), 500

@app.route('/api/chat', methods=['POST'])
def answer_question():
    try:
        start = time.time()
        data = request.get_json()
        
        if not data or 'question' not in data:
            return jsonify({'error': 'Missing "question" in request body'}), 400
        
        question = data['question']
        logger.info(f"📥 Chat request: {question[:50]}")
        
        model, tokenizer = model_cache.get_chat_model()
        
        input_text = f"answer question: {question}"
        input_ids = tokenizer.encode(input_text, return_tensors="pt").to(DEVICE)
        
        with torch.no_grad():
            outputs = model.generate(input_ids, max_length=256, num_beams=4)
        
        answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
        elapsed = time.time() - start
        
        logger.info(f"✅ Chat answer ({elapsed:.2f}s)")
        
        return jsonify({
            'question': question,
            'answer': answer,
            'inference_time': f'{elapsed:.3f}s',
            'model': 'chat-support'
        })
        
    except Exception as e:
        logger.error(f"❌ Chat error: {str(e)}")
        return jsonify({
            'error': 'Inference failed',
            'message': str(e)[:300]
        }), 500

@app.route('/api/service', methods=['POST'])
def generate_description():
    try:
        start = time.time()
        data = request.get_json()
        
        if not data or 'service_info' not in data:
            return jsonify({'error': 'Missing "service_info" in request body'}), 400
        
        service_info = data['service_info']
        logger.info(f"📥 Service description request")
        
        model, tokenizer = model_cache.get_service_model()
        
        input_text = f"generate professional description: {service_info}"
        input_ids = tokenizer.encode(input_text, return_tensors="pt").to(DEVICE)
        
        with torch.no_grad():
            outputs = model.generate(input_ids, max_length=512, num_beams=4)
        
        description = tokenizer.decode(outputs[0], skip_special_tokens=True)
        elapsed = time.time() - start
        
        logger.info(f"✅ Service description ({elapsed:.2f}s)")
        
        return jsonify({
            'service_info': service_info,
            'description': description,
            'inference_time': f'{elapsed:.3f}s',
            'model': 'service-description'
        })
        
    except Exception as e:
        logger.error(f"❌ Service error: {str(e)}")
        return jsonify({
            'error': 'Inference failed',
            'message': str(e)[:300]
        }), 500

@app.route('/api/search', methods=['POST'])
def semantic_search():
    try:
        start = time.time()
        data = request.get_json()
        
        if not data or 'query' not in data or 'documents' not in data:
            return jsonify({'error': 'Missing "query" or "documents"'}), 400
        
        query = data['query']
        documents = data['documents']
        
        if not isinstance(documents, list):
            return jsonify({'error': '"documents" must be a list'}), 400
        
        logger.info(f"📥 Search request: {query[:50]}")
        
        model, _ = model_cache.get_search_model()
        
        query_embedding = model.encode([query])
        doc_embeddings = model.encode(documents)
        
        from sklearn.metrics.pairwise import cosine_similarity
        similarities = cosine_similarity(query_embedding, doc_embeddings)[0]
        
        results = [
            {'document': doc, 'similarity': float(score), 'rank': i + 1}
            for i, (doc, score) in enumerate(
                sorted(zip(documents, similarities), key=lambda x: x[1], reverse=True)
            )
        ]
        
        elapsed = time.time() - start
        logger.info(f"✅ Search complete ({elapsed:.2f}s)")
        
        return jsonify({
            'query': query,
            'results': results,
            'inference_time': f'{elapsed:.3f}s',
            'model': 'semantic-search'
        })
        
    except Exception as e:
        logger.error(f"❌ Search error: {str(e)}")
        return jsonify({
            'error': 'Inference failed',
            'message': str(e)[:300]
        }), 500

@app.route('/api/moderate', methods=['POST'])
def moderate_content():
    try:
        start = time.time()
        data = request.get_json()
        
        if not data or 'text' not in data:
            return jsonify({'error': 'Missing "text" in request body'}), 400
        
        text = data['text']
        logger.info(f"📥 Moderation request")
        
        model, tokenizer = model_cache.get_moderation_model()
        
        inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
        inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
        
        with torch.no_grad():
            import torch.nn.functional as F
            outputs = model(**inputs)
            probs = F.softmax(outputs.logits, dim=-1)
            toxic_prob = probs[0][1].item()
        
        is_toxic = toxic_prob > 0.5
        elapsed = time.time() - start
        
        logger.info(f"✅ Moderation: {'toxic' if is_toxic else 'safe'} ({elapsed:.2f}s)")
        
        return jsonify({
            'text': text,
            'is_toxic': is_toxic,
            'toxicity_score': round(toxic_prob, 4),
            'status': 'toxic' if is_toxic else 'safe',
            'inference_time': f'{elapsed:.3f}s',
            'model': 'content-moderation'
        })
        
    except Exception as e:
        logger.error(f"❌ Moderation error: {str(e)}")
        return jsonify({
            'error': 'Inference failed',
            'message': str(e)[:300]
        }), 500

@app.errorhandler(404)
def not_found(error):
    return jsonify({'error': 'Endpoint not found', 'available': ['/', '/health', '/api/category', '/api/chat', '/api/service', '/api/search', '/api/moderate']}), 404

@app.errorhandler(500)
def internal_error(error):
    return jsonify({'error': 'Internal server error'}), 500

if __name__ == '__main__':
    port = int(os.getenv('PORT', 7860))
    logger.info(f"🚀 Starting NEED AI API on port {port}")
    logger.info(f"👤 Username: {HF_USERNAME}")
    logger.info(f"💡 Models will load on first request (30-60s each)")
    app.run(host='0.0.0.0', port=port, debug=False)