Spaces:

sixfingerdev
/

sixfinger-api

Running

App Files Files Community

sixfingerdev commited on Dec 18, 2025

Commit

b3e1f6a

verified ·

1 Parent(s): c314317

Update app.py

Browse files

Files changed (1) hide show

app.py +976 -579

app.py CHANGED Viewed

@@ -1,622 +1,1019 @@
-# app.py - Sixfinger Groq Backend (Allam-2-7B Eklendi)
-import json
-import os
-from datetime import datetime
-from flask import Flask, request, jsonify, Response
-from groq import Groq
-import traceback
 app = Flask(__name__)
-app.config['JSON_AS_ASCII'] = False
-# ========== CONFIGURATION ==========
-GROQ_API_KEY = os.getenv("GROQ_API_KEY")
-PORT = int(os.getenv("PORT", 7860))
-if not GROQ_API_KEY:
-    raise ValueError("❌ GROQ_API_KEY environment variable gerekli!")
-groq_client = Groq(api_key=GROQ_API_KEY)
-# ========== MODEL CATEGORIES ==========
-# FREE PLAN MODELS (Yüksek limitli)
-FREE_MODELS = {
-    'llama-8b-instant': {
-        'id': 'llama-3.1-8b-instant',
-        'description': 'Llama 3.1 8B Instant (Ultra Fast)',
-        'rpm': 30,
-        'rpd': 14400,  # ⭐ EN YÜKSEK
-        'tpm': 14400,
-        'tpd': 6000000,
-        'size': '8B',
-        'speed': '⚡⚡⚡',
-        'plan_required': 'free',
-        'language': 'Multilingual'
-    },
-    'allam-2-7b': {
-        'id': 'allam-2-7b',
-        'description': 'Allam 2 7B (Arabic/Turkish Optimized)',  # ✅ EKLENDI
-        'rpm': 30,
-        'rpd': 300,
-        'tpm': 7000,
-        'tpd': 60000,
-        'size': '7B',
-        'speed': '⚡⚡',
-        'plan_required': 'free',
-        'language': 'Arabic/Turkish'
-    }
 }
-# PAID PLAN MODELS (Güçlü ama düşük limit)
-PAID_MODELS = {
-    'llama-70b': {
-        'id': 'llama-3.3-70b-versatile',
-        'description': 'Llama 3.3 70B Versatile (Powerful)',
-        'rpm': 30,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 12000000,
-        'size': '70B',
-        'speed': '⚡⚡',
-        'plan_required': 'starter',
-        'language': 'Multilingual'
-    },
-    'qwen3-32b': {
-        'id': 'qwen/qwen3-32b',
-        'description': 'Qwen3 32B (Türkçe Optimized)',
-        'rpm': 60,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 6000000,
-        'size': '32B',
-        'speed': '⚡⚡',
-        'plan_required': 'starter',
-        'language': 'Turkish/Chinese'
-    },
-    'gpt-oss-120b': {
-        'id': 'openai/gpt-oss-120b',
-        'description': 'GPT OSS 120B (Giant)',
-        'rpm': 30,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 8000000,
-        'size': '120B',
-        'speed': '⚡⚡',
-        'plan_required': 'pro',
-        'language': 'Multilingual'
-    },
-    'llama-maverick-17b': {
-        'id': 'meta-llama/llama-4-maverick-17b-128e-instruct',
-        'description': 'Llama 4 Maverick 17B (Latest)',
-        'rpm': 30,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 6000000,
-        'size': '17B',
-        'speed': '⚡⚡',
-        'plan_required': 'starter',
-        'language': 'Multilingual'
-    },
-    'llama-scout-17b': {
-        'id': 'meta-llama/llama-4-scout-17b-16e-instruct',
-        'description': 'Llama 4 Scout 17B (Fast)',
-        'rpm': 30,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 30000000,  # Çok yüksek token limit!
-        'size': '17B',
-        'speed': '⚡⚡⚡',
-        'plan_required': 'starter',
-        'language': 'Multilingual'
-    },
-    'gpt-oss-20b': {
-        'id': 'openai/gpt-oss-20b',
-        'description': 'GPT OSS 20B (Compact)',
-        'rpm': 30,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 8000000,
-        'size': '20B',
-        'speed': '⚡⚡',
-        'plan_required': 'starter',
-        'language': 'Multilingual'
-    },
-    'kimi-k2': {
-        'id': 'moonshotai/kimi-k2-instruct',
-        'description': 'Kimi K2 Instruct (Chinese)',
-        'rpm': 60,
-        'rpd': 1000,
-        'tpm': 1000,
-        'tpd': 10000000,
-        'size': 'Unknown',
-        'speed': '⚡⚡',
-        'plan_required': 'pro',
-        'language': 'Chinese/Multilingual'
-    }
 }
-# TÜM MODELLER
-ALL_MODELS = {**FREE_MODELS, **PAID_MODELS}
-# DEFAULT MODEL PRIORITY (fallback için)
-MODEL_PRIORITY = [
-    # FREE (önce en yüksek limitli)
-    'llama-8b-instant',      # 14,400 RPD (FREE için ana)
-    'allam-2-7b',            # 300 RPD (FREE için yedek)
-    # PAID (güçlüden zayıfa)
-    'llama-70b',             # 70B (en güçlü genel amaçlı)
-    'gpt-oss-120b',          # 120B (giant)
-    'qwen3-32b',             # 32B (Türkçe)
-    'llama-scout-17b',       # 17B (hızlı + yüksek token limit)
-    'llama-maverick-17b',    # 17B (son model)
-    'gpt-oss-20b',           # 20B
-    'kimi-k2'                # Chinese
-]
-# ========== PLAN - MODEL MAPPING ==========
-PLAN_ALLOWED_MODELS = {
-    'free': [
-        'llama-8b-instant',  # Ana model (14.4K/gün)
-        'allam-2-7b'         # Yedek/alternatif (300/gün)
-    ],
-    'starter': [
-        'llama-8b-instant',
-        'allam-2-7b',
-        'qwen3-32b',         # Türkçe için
-        'llama-70b',         # Güçlü model
-        'llama-maverick-17b',
-        'llama-scout-17b',
-        'gpt-oss-20b'
-    ],
-    'pro': [
-        'llama-8b-instant',
-        'allam-2-7b',
-        'qwen3-32b',
-        'llama-70b',
-        'llama-maverick-17b',
-        'llama-scout-17b',
-        'gpt-oss-20b',
-        'gpt-oss-120b',      # Giant model
-        'kimi-k2'            # Chinese model
-    ],
-    'plus': list(ALL_MODELS.keys())  # Tüm modeller
-}
-# ========== STATISTICS ==========
-stats = {
-    'total_requests': 0,
-    'successful_requests': 0,
-    'failed_requests': 0,
-    'model_usage': {},
-    'model_failures': {},
-    'fallback_count': 0,
-    'start_time': datetime.utcnow()
-}
-# ========== HELPER FUNCTIONS ==========
-def get_allowed_models(user_plan='free', preferred_model=None):
-    """Kullanıcının planına göre izinli modelleri döndür"""
-    allowed = PLAN_ALLOWED_MODELS.get(user_plan, ['llama-8b-instant'])
-    # Preferred model varsa ve izinliyse öncelikli yap
-    if preferred_model and preferred_model in allowed:
-        models = [preferred_model] + [m for m in allowed if m != preferred_model]
-    else:
-        # MODEL_PRIORITY'ye göre sırala
-        models = [m for m in MODEL_PRIORITY if m in allowed]
-    return models
-def try_model(model_key, messages, max_tokens, temperature, top_p, stream=False):
-    """Bir model'i dene"""
-    if model_key not in ALL_MODELS:
-        return None, f"Unknown model: {model_key}"
-    model_info = ALL_MODELS[model_key]
-    model_id = model_info['id']
-    try:
-        if stream:
-            response = groq_client.chat.completions.create(
-                model=model_id,
-                messages=messages,
-                max_tokens=max_tokens,
-                temperature=temperature,
-                top_p=top_p,
-                stream=True
-            )
-            stats['model_usage'][model_key] = stats['model_usage'].get(model_key, 0) + 1
-            return response, None
-        else:
-            response = groq_client.chat.completions.create(
-                model=model_id,
-                messages=messages,
-                max_tokens=max_tokens,
-                temperature=temperature,
-                top_p=top_p
-            )
-            stats['model_usage'][model_key] = stats['model_usage'].get(model_key, 0) + 1
-            return response, None
-    except Exception as e:
-        error_msg = str(e)
-        stats['model_failures'][model_key] = stats['model_failures'].get(model_key, 0) + 1
-        print(f"❌ Model {model_key} failed: {error_msg}")
-        if 'rate_limit' in error_msg.lower() or 'rate limit' in error_msg.lower():
-            return None, "Rate limit exceeded"
-        elif 'quota' in error_msg.lower():
-            return None, "Quota exceeded"
-        elif 'timeout' in error_msg.lower():
-            return None, "Timeout"
-        else:
-            return None, f"Error: {error_msg[:150]}"
-def format_messages(prompt, system_prompt=None, history=None):
-    """Format messages"""
-    messages = []
-    if system_prompt:
-        messages.append({"role": "system", "content": system_prompt})
-    else:
-        # Default system prompt
-        messages.append({
-            "role": "system",
-            "content": "Sen yardımsever ve bilgili bir AI asistanısın. Türkçe'yi mükemmel kullanırsın."
-        })
-    if history:
-        messages.extend(history)
-    messages.append({"role": "user", "content": prompt})
-    return messages
-# ========== ROUTES ==========
-@app.route('/')
-def index():
-    """API Documentation"""
-    uptime = datetime.utcnow() - stats['start_time']
-    uptime_str = str(uptime).split('.')[0]
-    return jsonify({
-        'name': 'Sixfinger Groq Backend',
-        'version': '4.0.1',
-        'status': 'online',
-        'provider': 'Groq',
-        'uptime': uptime_str,
-        'models': {
-            'free': [
-                {'key': k, 'rpd': v['rpd'], 'language': v['language']}
-                for k, v in FREE_MODELS.items()
-            ],
-            'paid': [
-                {'key': k, 'rpd': v['rpd'], 'plan': v['plan_required']}
-                for k, v in PAID_MODELS.items()
-            ],
-            'total': len(ALL_MODELS)
-        },
-        'stats': {
-            'total_requests': stats['total_requests'],
-            'successful': stats['successful_requests'],
-            'failed': stats['failed_requests'],
-            'success_rate': f"{(stats['successful_requests'] / max(stats['total_requests'], 1) * 100):.2f}%",
-            'fallback_count': stats['fallback_count']
-        },
-        'endpoints': {
-            'chat': 'POST /api/chat',
-            'chat_stream': 'POST /api/chat/stream',
-            'models': 'GET /api/models',
-            'stats': 'GET /api/stats',
-            'health': 'GET /health'
-        },
-        'headers': {
-            'X-Model': 'Preferred model key (optional)',
-            'X-User-Plan': 'User plan: free, starter, pro, plus (default: free)'
-        }
-    })
-@app.route('/api/models')
-def list_models():
-    """List all models with details"""
-    return jsonify({
-        'free_models': [
-            {
-                'key': key,
-                'model_id': info['id'],
-                'description': info['description'],
-                'size': info['size'],
-                'speed': info['speed'],
-                'language': info['language'],
-                'limits': {
-                    'rpm': info['rpm'],
-                    'rpd': info['rpd'],
-                    'tpm': info['tpm'],
-                    'tpd': info['tpd']
-                },
-                'usage_count': stats['model_usage'].get(key, 0),
-                'failure_count': stats['model_failures'].get(key, 0)
             }
-            for key, info in FREE_MODELS.items()
-        ],
-        'paid_models': [
-            {
-                'key': key,
-                'model_id': info['id'],
-                'description': info['description'],
-                'size': info['size'],
-                'speed': info['speed'],
-                'language': info['language'],
-                'plan_required': info['plan_required'],
-                'limits': {
-                    'rpm': info['rpm'],
-                    'rpd': info['rpd'],
-                    'tpm': info['tpm'],
-                    'tpd': info['tpd']
-                },
-                'usage_count': stats['model_usage'].get(key, 0),
-                'failure_count': stats['model_failures'].get(key, 0)
             }
-            for key, info in PAID_MODELS.items()
-        ],
-        'plan_permissions': PLAN_ALLOWED_MODELS
-    })
-@app.route('/api/chat', methods=['POST'])
-def chat():
-    """Chat endpoint (non-streaming)"""
-    stats['total_requests'] += 1
-    try:
-        data = request.json
-        if not data or 'prompt' not in data:
-            stats['failed_requests'] += 1
-            return jsonify({'error': 'prompt required'}), 400
-        # Request parameters
-        prompt = data['prompt']
-        max_tokens = min(data.get('max_tokens', 1000), 4000)
-        temperature = min(max(data.get('temperature', 0.7), 0.0), 2.0)
-        top_p = min(max(data.get('top_p', 0.9), 0.1), 1.0)
-        system_prompt = data.get('system_prompt')
-        history = data.get('history', [])
-        # Model selection
-        preferred_model = request.headers.get('X-Model') or data.get('model')
-        user_plan = request.headers.get('X-User-Plan', 'free').lower()
-        # Validate plan
-        if user_plan not in PLAN_ALLOWED_MODELS:
-            user_plan = 'free'
-        # Get allowed models
-        models_to_try = get_allowed_models(user_plan, preferred_model)
-        # Format messages
-        messages = format_messages(prompt, system_prompt, history)
-        # Try models
-        attempts = []
-        for i, model_key in enumerate(models_to_try):
-            if i >= 5:  # Max 5 attempts
-                break
-            print(f"🔄 Trying model {i+1}/{min(5, len(models_to_try))}: {model_key}")
-            response, error = try_model(model_key, messages, max_tokens, temperature, top_p)
-            attempts.append({
-                'model': model_key,
-                'success': response is not None,
-                'error': error
-            })
-            if response:
-                content = response.choices[0].message.content
-                model_info = ALL_MODELS[model_key]
-                stats['successful_requests'] += 1
-                if i > 0:
-                    stats['fallback_count'] += 1
-                result = {
-                    'response': content,
-                    'model': model_info['id'],
-                    'model_key': model_key,
-                    'model_size': model_info['size'],
-                    'model_language': model_info['language'],
-                    'attempts': i + 1,
-                    'usage': {
-                        'prompt_tokens': response.usage.prompt_tokens,
-                        'completion_tokens': response.usage.completion_tokens,
-                        'total_tokens': response.usage.total_tokens
-                    },
-                    'parameters': {
-                        'max_tokens': max_tokens,
-                        'temperature': temperature,
-                        'top_p': top_p
-                    }
-                }
-                if i > 0:
-                    result['fallback_attempts'] = attempts
-                return jsonify(result)
-            else:
-                print(f"❌ {model_key} failed: {error}")
-                continue
-        stats['failed_requests'] += 1
-        return jsonify({
-            'error': 'All models failed',
-            'attempts': attempts,
-            'user_plan': user_plan,
-            'models_tried': [a['model'] for a in attempts]
-        }), 503
-    except Exception as e:
-        stats['failed_requests'] += 1
-        return jsonify({
-            'error': str(e),
-            'traceback': traceback.format_exc()
-        }), 500
-@app.route('/api/chat/stream', methods=['POST'])
-def chat_stream():
-    """Chat endpoint (streaming)"""
-    stats['total_requests'] += 1
-    try:
-        data = request.json
-        if not data or 'prompt' not in data:
-            return jsonify({'error': 'prompt required'}), 400
-        # Request parameters
-        prompt = data['prompt']
-        max_tokens = min(data.get('max_tokens', 1000), 4000)
-        temperature = min(max(data.get('temperature', 0.7), 0.0), 2.0)
-        top_p = min(max(data.get('top_p', 0.9), 0.1), 1.0)
-        system_prompt = data.get('system_prompt')
-        history = data.get('history', [])
-        # Model selection
-        preferred_model = request.headers.get('X-Model') or data.get('model')
-        user_plan = request.headers.get('X-User-Plan', 'free').lower()
-        if user_plan not in PLAN_ALLOWED_MODELS:
-            user_plan = 'free'
-        # Get allowed models
-        models_to_try = get_allowed_models(user_plan, preferred_model)
-        # Format messages
-        messages = format_messages(prompt, system_prompt, history)
-        def generate():
-            for i, model_key in enumerate(models_to_try):
-                if i >= 5:
-                    break
-                yield f"data: {json.dumps({'info': f'Trying model: {model_key}'}, ensure_ascii=False)}\n\n"
-                stream_response, error = try_model(model_key, messages, max_tokens, temperature, top_p, stream=True)
-                if stream_response:
-                    try:
-                        for chunk in stream_response:
-                            if chunk.choices[0].delta.content:
-                                text = chunk.choices[0].delta.content
-                                yield f"data: {json.dumps({'text': text}, ensure_ascii=False)}\n\n"
-                        stats['successful_requests'] += 1
-                        if i > 0:
-                            stats['fallback_count'] += 1
-                        model_info = ALL_MODELS[model_key]
-                        yield f"data: {json.dumps({'done': True, 'model': model_info['id'], 'model_key': model_key, 'attempts': i+1})}\n\n"
-                        return
-                    except Exception as e:
-                        yield f"data: {json.dumps({'warning': f'Stream error: {str(e)}'}, ensure_ascii=False)}\n\n"
-                        continue
-                else:
-                    yield f"data: {json.dumps({'warning': f'{model_key} failed: {error}'}, ensure_ascii=False)}\n\n"
-                    continue
-            stats['failed_requests'] += 1
-            yield f"data: {json.dumps({'error': 'All models failed'})}\n\n"
-        return Response(generate(), mimetype='text/event-stream')
-    except Exception as e:
-        stats['failed_requests'] += 1
-        return jsonify({'error': str(e)}), 500
-@app.route('/health')
-def health():
-    """Health check"""
-    uptime = datetime.utcnow() - stats['start_time']
-    return jsonify({
-        'status': 'ok',
-        'uptime_seconds': int(uptime.total_seconds()),
-        'total_requests': stats['total_requests'],
-        'success_rate': f"{(stats['successful_requests'] / max(stats['total_requests'], 1) * 100):.2f}%",
-        'timestamp': datetime.utcnow().isoformat()
-    })
-@app.route('/api/stats')
-def api_stats():
-    """Detailed statistics"""
-    uptime = datetime.utcnow() - stats['start_time']
-    return jsonify({
-        'uptime_seconds': int(uptime.total_seconds()),
-        'total_requests': stats['total_requests'],
-        'successful_requests': stats['successful_requests'],
-        'failed_requests': stats['failed_requests'],
-        'success_rate': f"{(stats['successful_requests'] / max(stats['total_requests'], 1) * 100):.2f}%",
-        'fallback_count': stats['fallback_count'],
-        'model_usage': stats['model_usage'],
-        'model_failures': stats['model_failures'],
-        'timestamp': datetime.utcnow().isoformat()
-    })
-# Error handlers
-@app.errorhandler(404)
-def not_found(e):
-    return jsonify({'error': 'Not found'}), 404
-@app.errorhandler(500)
-def internal_error(e):
-    return jsonify({'error': 'Internal error', 'detail': str(e)}), 500
-# CORS
-@app.after_request
-def after_request(response):
-    response.headers.add('Access-Control-Allow-Origin', '*')
-    response.headers.add('Access-Control-Allow-Headers', 'Content-Type,X-API-Key,X-Model,X-User-Plan')
-    response.headers.add('Access-Control-Allow-Methods', 'GET,POST,OPTIONS')
-    return response
-# ========== MAIN ==========
-if __name__ == '__main__':
-    print("\n" + "=" * 70)
-    print("🚀 SIXFINGER GROQ BACKEND v4.0.1")
-    print("=" * 70)
-    print(f"✅ Groq API Key: {GROQ_API_KEY[:20]}...")
-    print(f"📡 Port: {PORT}")
-    print("=" * 70)
-    print("\n🆓 FREE PLAN MODELS:")
-    for key, info in FREE_MODELS.items():
-        print(f"  • {key}: {info['description']}")
-        print(f"    RPD: {info['rpd']:,} | TPD: {info['tpd']:,} | Language: {info['language']}")
-    print("\n💎 PAID PLAN MODELS:")
-    for key, info in PAID_MODELS.items():
-        print(f"  • {key}: {info['description']}")
-        print(f"    Plan: {info['plan_required']}+ | RPD: {info['rpd']:,} | Language: {info['language']}")
-    print("\n📊 PLAN PERMISSIONS:")
-    for plan, models in PLAN_ALLOWED_MODELS.items():
-        print(f"  • {plan.upper()}: {len(models)} modeller - {', '.join(models[:3])}...")
-    print("\n" + "=" * 70)
-    print("✅ Server ready!")
-    print("📖 API Docs: http://0.0.0.0:7860")
-    print("=" * 70 + "\n")
-    app.run(host='0.0.0.0', port=PORT, debug=False, threaded=True)

+# app.py
+# Flask tabanlı Kayra Türkçe Dil Modeli Test Uygulaması
+# sixfingerdev/kayra-1 ve sixfingerdev/kayra-1-exp modellerini test etmenizi sağlar.
+from flask import Flask, request, jsonify, render_template_string
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
 app = Flask(__name__)
+# ==================== CSS STİLLERİ ====================
+STYLES = """
+:root {
+    --primary: #2c3e50;
+    --secondary: #3498db;
+    --accent: #e74c3c;
+    --success: #27ae60;
+    --warning: #f39c12;
+    --light: #ecf0f1;
+    --dark: #1a252f;
+    --gray: #95a5a6;
+    --white: #ffffff;
+    --shadow: 0 4px 20px rgba(0,0,0,0.1);
+    --radius: 12px;
 }
+* {
+    margin: 0;
+    padding: 0;
+    box-sizing: border-box;
 }
+body {
+    font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
+    background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+    min-height: 100vh;
+    color: var(--dark);
+    line-height: 1.6;
+}
+.navbar {
+    background: var(--dark);
+    padding: 1rem 2rem;
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    box-shadow: var(--shadow);
+}
+.navbar h1 {
+    color: var(--white);
+    font-size: 1.5rem;
+    font-weight: 600;
+}
+.nav-links {
+    display: flex;
+    gap: 1.5rem;
+}
+.nav-links a {
+    color: var(--light);
+    text-decoration: none;
+    font-weight: 500;
+    padding: 0.5rem 1rem;
+    border-radius: 6px;
+    transition: all 0.3s ease;
+}
+.nav-links a:hover,
+.nav-links a.active {
+    background: var(--secondary);
+    color: var(--white);
+}
+.container {
+    max-width: 1000px;
+    margin: 2rem auto;
+    padding: 0 1rem;
+}
+.card {
+    background: var(--white);
+    border-radius: var(--radius);
+    box-shadow: var(--shadow);
+    padding: 2rem;
+    margin-bottom: 1.5rem;
+}
+.card-header {
+    border-bottom: 2px solid var(--light);
+    padding-bottom: 1rem;
+    margin-bottom: 1.5rem;
+}
+.card-header h2 {
+    color: var(--primary);
+    font-size: 1.5rem;
+}
+.card-header p {
+    color: var(--gray);
+    margin-top: 0.5rem;
+}
+label {
+    display: block;
+    font-weight: 600;
+    color: var(--primary);
+    margin-bottom: 0.5rem;
+}
+textarea {
+    width: 100%;
+    height: 140px;
+    padding: 1rem;
+    border: 2px solid var(--light);
+    border-radius: 8px;
+    font-size: 1rem;
+    font-family: inherit;
+    resize: vertical;
+    transition: border-color 0.3s ease;
+}
+textarea:focus {
+    outline: none;
+    border-color: var(--secondary);
+}
+select {
+    width: 100%;
+    padding: 0.8rem 1rem;
+    border: 2px solid var(--light);
+    border-radius: 8px;
+    font-size: 1rem;
+    font-family: inherit;
+    background: var(--white);
+    cursor: pointer;
+    margin-bottom: 1rem;
+}
+select:focus {
+    outline: none;
+    border-color: var(--secondary);
+}
+.btn {
+    display: inline-block;
+    padding: 1rem 2rem;
+    font-size: 1rem;
+    font-weight: 600;
+    border: none;
+    border-radius: 8px;
+    cursor: pointer;
+    transition: all 0.3s ease;
+    width: 100%;
+    text-align: center;
+}
+.btn-primary {
+    background: linear-gradient(135deg, var(--secondary), #2980b9);
+    color: var(--white);
+}
+.btn-primary:hover {
+    transform: translateY(-2px);
+    box-shadow: 0 6px 20px rgba(52, 152, 219, 0.4);
+}
+.btn-primary:disabled {
+    background: var(--gray);
+    cursor: not-allowed;
+    transform: none;
+}
+.form-group {
+    margin-bottom: 1.5rem;
+}
+.response-box {
+    background: linear-gradient(135deg, #f8f9fa, #e9ecef);
+    border-radius: var(--radius);
+    padding: 1.5rem;
+    margin-top: 1.5rem;
+    display: none;
+}
+.response-box.active {
+    display: block;
+}
+.response-header {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin-bottom: 1rem;
+    padding-bottom: 0.5rem;
+    border-bottom: 1px solid var(--gray);
+}
+.model-badge {
+    background: var(--secondary);
+    color: var(--white);
+    padding: 0.3rem 0.8rem;
+    border-radius: 20px;
+    font-size: 0.85rem;
+    font-weight: 600;
+}
+.response-content {
+    white-space: pre-wrap;
+    font-size: 1rem;
+}
+.response-content strong {
+    color: var(--primary);
+}
+.loading {
+    text-align: center;
+    padding: 2rem;
+    color: var(--gray);
+    display: none;
+}
+.loading.active {
+    display: block;
+}
+.spinner {
+    width: 40px;
+    height: 40px;
+    border: 4px solid var(--light);
+    border-top: 4px solid var(--secondary);
+    border-radius: 50%;
+    animation: spin 1s linear infinite;
+    margin: 0 auto 1rem;
+}
+@keyframes spin {
+    0% { transform: rotate(0deg); }
+    100% { transform: rotate(360deg); }
+}
+.error-box {
+    background: #fdeaea;
+    border: 1px solid var(--accent);
+    color: var(--accent);
+    padding: 1rem;
+    border-radius: 8px;
+    margin-top: 1rem;
+}
+.info-section {
+    margin-bottom: 2rem;
+}
+.info-section h3 {
+    color: var(--primary);
+    margin-bottom: 1rem;
+    padding-bottom: 0.5rem;
+    border-bottom: 2px solid var(--secondary);
+}
+.stats-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(200px, 1fr));
+    gap: 1rem;
+    margin: 1rem 0;
+}
+.stat-card {
+    background: linear-gradient(135deg, var(--light), #dfe6e9);
+    padding: 1.5rem;
+    border-radius: 8px;
+    text-align: center;
+}
+.stat-value {
+    font-size: 2rem;
+    font-weight: 700;
+    color: var(--secondary);
+}
+.stat-label {
+    color: var(--gray);
+    font-size: 0.9rem;
+    margin-top: 0.3rem;
+}
+table {
+    width: 100%;
+    border-collapse: collapse;
+    margin: 1rem 0;
+}
+th, td {
+    padding: 0.8rem;
+    text-align: left;
+    border-bottom: 1px solid var(--light);
+}
+th {
+    background: var(--primary);
+    color: var(--white);
+}
+tr:hover {
+    background: var(--light);
+}
+.badge {
+    display: inline-block;
+    padding: 0.2rem 0.6rem;
+    border-radius: 4px;
+    font-size: 0.8rem;
+    font-weight: 600;
+}
+.badge-success {
+    background: #d4edda;
+    color: var(--success);
+}
+.badge-danger {
+    background: #f8d7da;
+    color: var(--accent);
+}
+.badge-warning {
+    background: #fff3cd;
+    color: #856404;
+}
+.badge-info {
+    background: #d1ecf1;
+    color: #0c5460;
+}
+.code-block {
+    background: var(--dark);
+    color: #a6e22e;
+    padding: 1rem;
+    border-radius: 8px;
+    overflow-x: auto;
+    font-family: 'Consolas', 'Monaco', monospace;
+    font-size: 0.9rem;
+    margin: 1rem 0;
+}
+.highlight {
+    background: linear-gradient(135deg, #fff3cd, #ffeeba);
+    padding: 1rem;
+    border-radius: 8px;
+    border-left: 4px solid var(--warning);
+    margin: 1rem 0;
+}
+.comparison-table tr:nth-child(even) {
+    background: #f8f9fa;
+}
+.footer {
+    text-align: center;
+    padding: 2rem;
+    color: var(--white);
+    opacity: 0.8;
+}
+.footer a {
+    color: var(--white);
+}
+@media (max-width: 768px) {
+    .navbar {
+        flex-direction: column;
+        gap: 1rem;
+    }
+    .stats-grid {
+        grid-template-columns: 1fr;
+    }
+    .two-column {
+        grid-template-columns: 1fr;
+    }
+}
+.two-column {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 1rem;
+}
+"""
+# ==================== ANA SAYFA HTML ====================
+INDEX_HTML = """
+<!DOCTYPE html>
+<html lang="tr">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Kayra - Türkçe Dil Modeli Test</title>
+    <style>{{ styles }}</style>
+</head>
+<body>
+    <nav class="navbar">
+        <h1>Kayra Türkçe Dil Modeli</h1>
+        <div class="nav-links">
+            <a href="/" class="active">Test</a>
+            <a href="/info">Bilgi</a>
+        </div>
+    </nav>
+    <div class="container">
+        <div class="card">
+            <div class="card-header">
+                <h2>Model Test Arayüzü</h2>
+                <p>Küçük ölçekli Türkçe dil modellerini karşılaştırmalı olarak test edin.</p>
+            </div>
+            <div class="form-group">
+                <label for="prompt">Soru veya mesajınız:</label>
+                <textarea id="prompt" placeholder="Örnek: Türkiye'nin başkenti neresidir?"></textarea>
+            </div>
+            <div class="form-group">
+                <label for="model">Model seçin:</label>
+                <select id="model">
+                    <option value="stable">kayra-1 (Stable - Instruction Tuned)</option>
+                    <option value="exp">kayra-1-exp (Deneysel - Sadece Pretrained)</option>
+                </select>
+            </div>
+            <button class="btn btn-primary" id="submitBtn" onclick="generate()">
+                Gönder ve Cevap Al
+            </button>
+            <div class="loading" id="loading">
+                <div class="spinner"></div>
+                <p>Yanıt oluşturuluyor, lütfen bekleyin...</p>
+            </div>
+            <div class="response-box" id="responseBox">
+                <div class="response-header">
+                    <span>Model:</span>
+                    <span class="model-badge" id="modelBadge"></span>
+                </div>
+                <div class="response-content" id="responseContent"></div>
+            </div>
+            <div class="error-box" id="errorBox" style="display:none;"></div>
+        </div>
+    </div>
+    <div class="footer">
+        <p>Kayra - Sıfırdan Türkçe ile eğitilmiş GPT modeli</p>
+        <p><a href="https://huggingface.co/sixfingerdev" target="_blank">Hugging Face</a></p>
+    </div>
+    <script>
+        async function generate() {
+            const prompt = document.getElementById("prompt").value.trim();
+            const model = document.getElementById("model").value;
+            const responseBox = document.getElementById("responseBox");
+            const loading = document.getElementById("loading");
+            const errorBox = document.getElementById("errorBox");
+            const submitBtn = document.getElementById("submitBtn");
+            if (!prompt) {
+                errorBox.textContent = "Lütfen bir mesaj girin!";
+                errorBox.style.display = "block";
+                responseBox.classList.remove("active");
+                return;
             }
+            errorBox.style.display = "none";
+            responseBox.classList.remove("active");
+            loading.classList.add("active");
+            submitBtn.disabled = true;
+            try {
+                const response = await fetch("/generate", {
+                    method: "POST",
+                    headers: { "Content-Type": "application/json" },
+                    body: JSON.stringify({ prompt, model })
+                });
+                const data = await response.json();
+                if (data.error) {
+                    errorBox.textContent = "Hata: " + data.error;
+                    errorBox.style.display = "block";
+                } else {
+                    document.getElementById("modelBadge").textContent = data.model;
+                    document.getElementById("responseContent").innerHTML =
+                        "<strong>Soru:</strong> " + escapeHtml(data.prompt) +
+                        "\\n\\n<strong>Cevap:</strong>\\n" + escapeHtml(data.response);
+                    responseBox.classList.add("active");
+                }
+            } catch (err) {
+                errorBox.textContent = "Bağlantı hatası: " + err.message;
+                errorBox.style.display = "block";
+            } finally {
+                loading.classList.remove("active");
+                submitBtn.disabled = false;
             }
+        }
+        function escapeHtml(text) {
+            const div = document.createElement('div');
+            div.textContent = text;
+            return div.innerHTML;
+        }
+        document.getElementById("prompt").addEventListener("keydown", function(e) {
+            if (e.ctrlKey && e.key === "Enter") {
+                generate();
+            }
+        });
+    </script>
+</body>
+</html>
+"""
+# ==================== BİLGİ SAYFASI HTML ====================
+INFO_HTML = """
+<!DOCTYPE html>
+<html lang="tr">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>Kayra - Model Bilgileri</title>
+    <style>{{ styles }}</style>
+</head>
+<body>
+    <nav class="navbar">
+        <h1>Kayra Türkçe Dil Modeli</h1>
+        <div class="nav-links">
+            <a href="/">Test</a>
+            <a href="/info" class="active">Bilgi</a>
+        </div>
+    </nav>
+    <div class="container">
+        <!-- Genel Bakış -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Kayra Nedir?</h2>
+                <p>Sıfırdan Türkçe ile eğitilmiş deneysel GPT modelleri</p>
+            </div>
+            <p>Kayra, Türkçe dil işleme araştırmaları için geliştirilmiş, sıfırdan eğitilmiş
+            küçük ölçekli GPT tabanlı dil modelleridir. Bu projede iki farklı model bulunmaktadır:</p>
+            <div class="stats-grid">
+                <div class="stat-card">
+                    <div class="stat-value">85M</div>
+                    <div class="stat-label">Parametre</div>
+                </div>
+                <div class="stat-card">
+                    <div class="stat-value">500K</div>
+                    <div class="stat-label">Eğitim Dokümanı</div>
+                </div>
+                <div class="stat-card">
+                    <div class="stat-value">42.7</div>
+                    <div class="stat-label">Validation PPL</div>
+                </div>
+                <div class="stat-card">
+                    <div class="stat-value">MIT</div>
+                    <div class="stat-label">Lisans</div>
+                </div>
+            </div>
+        </div>
+        <!-- Model Karşılaştırması -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Model Karşılaştırması</h2>
+            </div>
+            <table>
+                <thead>
+                    <tr>
+                        <th>Özellik</th>
+                        <th>kayra-1 (Stable)</th>
+                        <th>kayra-1-exp (Deneysel)</th>
+                    </tr>
+                </thead>
+                <tbody>
+                    <tr>
+                        <td>Tür</td>
+                        <td>Instruction-tuned</td>
+                        <td>Sadece Pretrained</td>
+                    </tr>
+                    <tr>
+                        <td>Kullanım</td>
+                        <td>Soru-Cevap formatında</td>
+                        <td>Metin tamamlama</td>
+                    </tr>
+                    <tr>
+                        <td>Stabilite</td>
+                        <td><span class="badge badge-success">Stabil</span></td>
+                        <td><span class="badge badge-warning">Deneysel</span></td>
+                    </tr>
+                    <tr>
+                        <td>Prompt Formatı</td>
+                        <td>### Soru: ... ### Cevap:</td>
+                        <td>Düz metin</td>
+                    </tr>
+                    <tr>
+                        <td>Önerilen Kullanım</td>
+                        <td>Genel test ve demo</td>
+                        <td>Araştırma ve analiz</td>
+                    </tr>
+                </tbody>
+            </table>
+        </div>
+        <!-- Mimari Detaylar -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Teknik Mimari</h2>
+            </div>
+            <table>
+                <tbody>
+                    <tr><td><strong>Model Türü</strong></td><td>Decoder-only Transformer (GPT-style)</td></tr>
+                    <tr><td><strong>Katman Sayısı</strong></td><td>10</td></tr>
+                    <tr><td><strong>Hidden Size</strong></td><td>640</td></tr>
+                    <tr><td><strong>Attention Heads</strong></td><td>10</td></tr>
+                    <tr><td><strong>FFN Size</strong></td><td>2560</td></tr>
+                    <tr><td><strong>Vocabulary</strong></td><td>32,000 BPE tokens</td></tr>
+                    <tr><td><strong>Context Length</strong></td><td>512 tokens</td></tr>
+                    <tr><td><strong>Toplam Parametre</strong></td><td>~85 milyon</td></tr>
+                </tbody>
+            </table>
+        </div>
+        <!-- Eğitim Verisi -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Eğitim Verisi</h2>
+            </div>
+            <table>
+                <thead>
+                    <tr>
+                        <th>Kaynak</th>
+                        <th>Doküman Sayısı</th>
+                        <th>Açıklama</th>
+                    </tr>
+                </thead>
+                <tbody>
+                    <tr>
+                        <td>Wikipedia TR</td>
+                        <td>~170,000</td>
+                        <td>Türkçe Vikipedi makaleleri</td>
+                    </tr>
+                    <tr>
+                        <td>mC4 Turkish</td>
+                        <td>~330,000</td>
+                        <td>Common Crawl web dokümanları</td>
+                    </tr>
+                    <tr>
+                        <td><strong>Toplam</strong></td>
+                        <td><strong>~500,000</strong></td>
+                        <td>MinHash LSH ile dedupe edilmiş</td>
+                    </tr>
+                </tbody>
+            </table>
+        </div>
+        <!-- Hallucination Problemi -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Neden Küçük Modeller Yanlış Bilgi Üretir?</h2>
+            </div>
+            <div class="highlight">
+                <strong>Temel Bulgu:</strong> Loss azalması, gerçek bilgi doğruluğu anlamına gelmez.
+                (loss down != factual accuracy up)
+            </div>
+            <div class="info-section">
+                <h3>Eğitim Sürecinde Gözlemlenen Davranış</h3>
+                <table class="comparison-table">
+                    <thead>
+                        <tr>
+                            <th>Step</th>
+                            <th>Val Loss</th>
+                            <th>Val PPL</th>
+                            <th>Üretilen Başkent</th>
+                            <th>Doğru mu?</th>
+                        </tr>
+                    </thead>
+                    <tbody>
+                        <tr>
+                            <td>1000</td>
+                            <td>5.98</td>
+                            <td>397.3</td>
+                            <td>Ankara</td>
+                            <td><span class="badge badge-success">Evet</span></td>
+                        </tr>
+                        <tr>
+                            <td>3000</td>
+                            <td>3.94</td>
+                            <td>51.7</td>
+                            <td>Ankara</td>
+                            <td><span class="badge badge-success">Evet</span></td>
+                        </tr>
+                        <tr>
+                            <td>5000</td>
+                            <td>4.02</td>
+                            <td>56.2</td>
+                            <td>Rastgele şehir</td>
+                            <td><span class="badge badge-danger">Hayır</span></td>
+                        </tr>
+                        <tr>
+                            <td>6500</td>
+                            <td>3.90</td>
+                            <td>49.6</td>
+                            <td>Bolu</td>
+                            <td><span class="badge badge-danger">Hayır</span></td>
+                        </tr>
+                        <tr>
+                            <td>7500</td>
+                            <td>3.83</td>
+                            <td>46.1</td>
+                            <td>Konya</td>
+                            <td><span class="badge badge-danger">Hayır</span></td>
+                        </tr>
+                        <tr>
+                            <td>9000</td>
+                            <td>3.75</td>
+                            <td>42.7</td>
+                            <td>Ankara (bazen)</td>
+                            <td><span class="badge badge-warning">Belirsiz</span></td>
+                        </tr>
+                    </tbody>
+                </table>
+            </div>
+            <div class="two-column">
+                <div class="info-section">
+                    <h3>Modelin Başarıyla Öğrendikleri</h3>
+                    <ul style="margin-left: 1.5rem;">
+                        <li>Dilbilgisi (Türkçe morfoloji)</li>
+                        <li>Cümle yapısı (syntax)</li>
+                        <li>Stil (resmi/gündelik ton eşleştirme)</li>
+                        <li>Bağlam tutarlılığı (konu tutarlılığı)</li>
+                        <li>Örüntü eşleme (Wikipedia tarzında metin)</li>
+                    </ul>
+                </div>
+                <div class="info-section">
+                    <h3>Modelin Öğrenemedikleri</h3>
+                    <ul style="margin-left: 1.5rem;">
+                        <li>Gerçek temeli: "Ankara = başkent" deterministik kuralı</li>
+                        <li>Mantıksal tutarlılık: Aynı prompt aynı gerçeği vermeli</li>
+                        <li>Bilgi geri çağırma: Güvenilir bilgi hatırlaması</li>
+                        <li>Gerçek vs örüntü: Doğruluğu olabilirlikten ayırt etme</li>
+                    </ul>
+                </div>
+            </div>
+        </div>
+        <!-- Model Boyutu Karşılaştırması -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Model Boyutu ve Gerçeklik İlişkisi</h2>
+            </div>
+            <table>
+                <thead>
+                    <tr>
+                        <th>Model</th>
+                        <th>Parametre</th>
+                        <th>Gerçeklik Güvenilirliği</th>
+                    </tr>
+                </thead>
+                <tbody>
+                    <tr>
+                        <td>Kayra (bu model)</td>
+                        <td>85M</td>
+                        <td><span class="badge badge-danger">Düşük - hallucination yaygın</span></td>
+                    </tr>
+                    <tr>
+                        <td>GPT-2 Small</td>
+                        <td>124M</td>
+                        <td><span class="badge badge-danger">Düşük - benzer sorunlar</span></td>
+                    </tr>
+                    <tr>
+                        <td>GPT-2 Medium</td>
+                        <td>355M</td>
+                        <td><span class="badge badge-warning">Orta - hala güvenilmez</span></td>
+                    </tr>
+                    <tr>
+                        <td>GPT-3</td>
+                        <td>175B</td>
+                        <td><span class="badge badge-success">İyi tutarlılık</span></td>
+                    </tr>
+                    <tr>
+                        <td>GPT-4</td>
+                        <td>~1.7T + RLHF</td>
+                        <td><span class="badge badge-success">Güvenilir</span></td>
+                    </tr>
+                    <tr>
+                        <td>GPT-5</td>
+                        <td>~10T (tahmini)</td>
+                        <td><span class="badge badge-info">Çok yüksek (beklenen)</span></td>
+                    </tr>
+                </tbody>
+            </table>
+            <div class="highlight">
+                <strong>Sonuç:</strong> 85M parametre dil örüntülerini öğrenir, bilgi tabanı oluşturmaz.
+                Gerçek bilgi güvenilirliği için milyarlarca parametre ve RLHF/DPO gibi hizalama teknikleri gerekir.
+            </div>
+        </div>
+        <!-- Kullanım Örneği -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Kod Örneği</h2>
+            </div>
+            <div class="code-block">
+<pre>from transformers import AutoModelForCausalLM, AutoTokenizer
+model = AutoModelForCausalLM.from_pretrained(
+    "sixfingerdev/kayra-1-exp",
+    trust_remote_code=True  # ONEMLI!
+)
+tokenizer = AutoTokenizer.from_pretrained("sixfingerdev/kayra-1-exp")
+prompt = "Türkiye'nin başkenti"
+inputs = tokenizer(prompt, return_tensors="pt")
+outputs = model.generate(
+    inputs.input_ids,
+    max_new_tokens=100,
+    temperature=0.8,
+    top_k=50,
+    repetition_penalty=1.2,
+    do_sample=True
+)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))</pre>
+            </div>
+        </div>
+        <!-- Uygun Kullanım Alanları -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Kullanım Alanları</h2>
+            </div>
+            <div class="two-column">
+                <div>
+                    <h4 style="color: var(--success); margin-bottom: 0.5rem;">Önerilen Kullanım</h4>
+                    <ul style="margin-left: 1.5rem;">
+                        <li>Türkçe NLP limitasyonları araştırması</li>
+                        <li>Pretraining baseline karşılaştırmaları</li>
+                        <li>Hallucination örüntü çalışmaları</li>
+                        <li>Eğitim amaçlı demonstrasyonlar</li>
+                        <li>LLM başarısızlık modlarını anlama</li>
+                    </ul>
+                </div>
+                <div>
+                    <h4 style="color: var(--accent); margin-bottom: 0.5rem;">Önerilmeyen Kullanım</h4>
+                    <ul style="margin-left: 1.5rem;">
+                        <li>Production uygulamaları</li>
+                        <li>Gerçek soru-cevap sistemleri</li>
+                        <li>Bilgi erişim sistemleri</li>
+                        <li>Eğitim içeriği üretimi</li>
+                        <li>Doğruluk gerektiren her iş</li>
+                    </ul>
+                </div>
+            </div>
+        </div>
+        <!-- Gelecek Planları -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Gelecek: Kayra-v2</h2>
+            </div>
+            <ul style="margin-left: 1.5rem;">
+                <li><strong>Daha büyük model:</strong> 350M-750M parametre</li>
+                <li><strong>Daha iyi tokenizer:</strong> NFC Unicode normalization</li>
+                <li><strong>Instruction tuning:</strong> 10K doğrulanmış QA çifti</li>
+                <li><strong>Alignment:</strong> Gerçeklik için RLHF veya DPO</li>
+                <li><strong>Değerlendirme:</strong> Gerçek kontrol benchmark'ları</li>
+            </ul>
+        </div>
+        <!-- Lisans ve Atıf -->
+        <div class="card">
+            <div class="card-header">
+                <h2>Lisans ve Atıf</h2>
+            </div>
+            <p><strong>Lisans:</strong> MIT License - Ticari ve akademik kullanım serbesttir.</p>
+            <div class="code-block">
+<pre>@misc{kayra2024hallucination,
+  title={Why Small Turkish GPTs Hallucinate Facts: An Experimental 85M Model},
+  author={sixfingerdev},
+  year={2024},
+  publisher={HuggingFace},
+  howpublished={\\url{https://huggingface.co/sixfingerdev/kayra-1-exp}},
+  note={Research on loss-factuality divergence in low-resource language models}
+}</pre>
+            </div>
+            <div class="highlight" style="margin-top: 1rem;">
+                <strong>Uyarı:</strong> Bu model bilerek kusurlarıyla birlikte paylaşılmıştır.
+                Küçük LM'lerin neden hallucination yaptığını gösteren bir öğrenme kaynağı olarak
+                hizmet eder, production aracı olarak değil.
+            </div>
+        </div>
+    </div>
+    <div class="footer">
+        <p>Kayra - Türkçe'yi Yaratan Zeka</p>
+        <p><a href="https://huggingface.co/sixfingerdev" target="_blank">Hugging Face</a></p>
+    </div>
+</body>
+</html>
+"""
+# ==================== MODEL YÜKLEME ====================
+print("Modeller yükleniyor... Bu biraz sürebilir (özellikle ilk seferde).")
+model_stable = AutoModelForCausalLM.from_pretrained(
+    "sixfingerdev/kayra-1",
+    trust_remote_code=True,
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto"
+)
+tokenizer_stable = AutoTokenizer.from_pretrained("sixfingerdev/kayra-1")
+model_exp = AutoModelForCausalLM.from_pretrained(
+    "sixfingerdev/kayra-1-exp",
+    trust_remote_code=True,
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto"
+)
+tokenizer_exp = AutoTokenizer.from_pretrained("sixfingerdev/kayra-1-exp")
+print("Modeller başarıyla yüklendi!")
+# ==================== YARDIMCI FONKSİYONLAR ====================
+def generate_response(model, tokenizer, prompt, max_new_tokens=150):
+    if model == model_stable:
+        formatted_prompt = f"### Soru: {prompt}\n\n### Cevap:"
+    else:
+        formatted_prompt = prompt
+    inputs = tokenizer(formatted_prompt, return_tensors="pt")
+    inputs = inputs.to(model.device)
+    if "token_type_ids" in inputs:
+        inputs.pop("token_type_ids")
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_new_tokens,
+            temperature=0.7,
+            do_sample=True,
+            top_p=0.9,
+            repetition_penalty=1.3,
+            pad_token_id=tokenizer.eos_token_id
+        )
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    if model == model_stable:
+        if "### Cevap:" in response:
+            response = response.split("### Cevap:")[-1].strip()
+    return response
+# ==================== ROUTE'LAR ====================
+@app.route("/")
+def index():
+    return render_template_string(INDEX_HTML, styles=STYLES)
+@app.route("/info")
+def info():
+    return render_template_string(INFO_HTML, styles=STYLES)
+@app.route("/generate", methods=["POST"])
+def generate():
+    data = request.json
+    prompt = data.get("prompt", "").strip()
+    model_choice = data.get("model", "stable")
+    if not prompt:
+        return jsonify({"error": "Lütfen bir soru veya mesaj girin."})
+    if model_choice == "stable":
+        response = generate_response(model_stable, tokenizer_stable, prompt)
+        model_name = "kayra-1 (Stable)"
+    else:
+        response = generate_response(model_exp, tokenizer_exp, prompt)
+        model_name = "kayra-1-exp (Deneysel)"
+    return jsonify({
+        "model": model_name,
+        "prompt": prompt,
+        "response": response
+    })
+# ==================== UYGULAMA BAŞLATMA ====================
+if __name__ == "__main__":
+    app.run(host="0.0.0.0", port=5000, debug=True)