Add Damascus Real Estate Falcon 7B test

README.md

@@ -1,12 +1,54 @@
 ---
-title: Falcon Testing
-emoji: 🌖
-colorFrom: pink
-colorTo: indigo
+title: Damascus Real Estate Chatbot
+emoji: 🏠
+colorFrom: blue
+colorTo: green
 sdk: gradio
-sdk_version: 5.40.0
+sdk_version: 4.0.2
 app_file: app.py
 pinned: false
+license: mit
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Damascus Real Estate Chatbot - Falcon 7B Test
+
+Testing Falcon 7B Instruct for Syrian Arabic real estate queries on Hugging Face Spaces.
+
+## Overview
+
+This is a test deployment of the Damascus Real Estate Chatbot using:
+- **Falcon 7B Instruct** for Arabic language understanding
+- **Syrian Arabic dialect** for natural conversation
+- **20 Damascus property listings** for real estate matching
+- **Entity extraction** for user requirements
+- **Gradio interface** for easy testing
+
+## Features
+
+- 🤖 **Falcon 7B Instruct** - Advanced Arabic language model
+- 🏠 **Real Estate Matching** - Finds properties based on user requirements
+- 💬 **Syrian Dialect** - Natural Damascus conversational style
+- 📋 **Entity Extraction** - Understands location, rooms, features
+- 🔍 **Smart Filtering** - Matches properties to user criteria
+
+## Usage
+
+Ask questions in Syrian Arabic like:
+- "بدي شقة مفروشة بالمالكي مع شرفة"
+- "في شي ستوديو باب شرقي مفروش؟"
+- "شو عندك بالشعلان رخيص؟"
+
+## Next Steps
+
+After testing on Hugging Face Spaces, this will be deployed to:
+- **AWS EC2 G4 instances** for production
+- **vLLM integration** for parallel inference
+- **ChromaDB RAG** for semantic search
+- **Performance optimization** for real-world usage
+
+## Technical Details
+
+- **Model**: tiiuae/falcon-7b-instruct
+- **Framework**: Transformers + Gradio
+- **Languages**: Syrian Arabic, some English for interface
+- **Data**: 20 Damascus property listings

app.py

@@ -0,0 +1,301 @@
+"""
+Falcon 7B Damascus Real Estate Chatbot - Hugging Face Test
+==========================================================
+
+Testing Falcon 7B Instruct on Hugging Face Spaces for Syrian Arabic real estate queries.
+This is the base test before deploying to AWS EC2 G4.
+"""
+
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+import re
+import json
+from typing import List, Dict, Any
+
+# Configuration
+MODEL_NAME = "tiiuae/falcon-7b-instruct"
+MAX_LENGTH = 500
+TEMPERATURE = 0.7
+
+print("🚀 Loading Falcon 7B for Damascus Real Estate...")
+
+# Load model and tokenizer
+try:
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_NAME,
+        torch_dtype=torch.float16,
+        device_map="auto",
+        trust_remote_code=True
+    )
+    
+    generator = pipeline(
+        "text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        torch_dtype=torch.float16,
+        device=0 if torch.cuda.is_available() else -1
+    )
+    
+    print("✅ Falcon 7B loaded successfully")
+    model_loaded = True
+    
+except Exception as e:
+    print(f"❌ Model loading failed: {e}")
+    generator = None
+    model_loaded = False
+
+# Damascus property listings (from Project_Description.txt)
+DAMASCUS_LISTINGS = [
+    "1. مفروشة بالكامل، غرفتين وصالون، المالكي، طابق ثالث، شرفة واسعة، غسالة صحون، مليون وثمانمائة ألف ليرة بالشهر.",
+    "2. ستوديو مفروش، باب شرقي، طابق أول، شرفة صغيرة، مكيف، ثمانمائة ألف ليرة بالشهر.",
+    "3. شقة فارغة، ثلاث غرف، المزة، طابق ثاني، شرفة كبيرة، موقف سيارة، مليونين وخمسمائة ألف ليرة بالشهر.",
+    "4. مفروشة، غرفة وصالون، الشعلان، طابق رابع، مكيفات، غسالة ملابس، مليون ومائتا ألف ليرة بالشهر.",
+    "5. دوبلكس فاخر، أربع غرف، أبو رمانة، طابقين، تراس كبير، مصعد، خمسة ملايين ليرة بالشهر.",
+    "6. ستوديو فارغ، القصاع، طابق أرضي، بدون شرفة، خمسمائة ألف ليرة بالشهر.",
+    "7. مفروشة جزئياً، غرفتين، الجرمانا، طابق ثالث، شرفة متوسطة، غسالة صحون، مليون وأربعمائة ألف ليرة بالشهر.",
+    "8. شقة فارغة، غرفة وصالون، باب توما، طابق ثاني، شرفة صغيرة، تسعمائة ألف ليرة بالشهر.",
+    "9. مفروشة بالكامل، ثلاث غرف، المالكي، طابق أول، شرفتين، مكيفات، موقف سيارة، ثلاثة ملايين ليرة بالشهر.",
+    "10. ستوديو مفروش، الشعلان، طابق خامس، شرفة واسعة، مكيف، فرن كهربائي، مليون ليرة بالشهر."
+]
+
+def create_falcon_prompt(user_query: str, listings: List[str]) -> str:
+    """Create a prompt for Falcon 7B to respond in Syrian Arabic"""
+    
+    listings_text = "\n".join(listings)
+    
+    prompt = f"""You are a real estate agent in Damascus, Syria. You help people find apartments.
+
+Available Properties:
+{listings_text}
+
+User Query: "{user_query}"
+
+Instructions:
+1. Respond ONLY in Syrian Arabic (Damascus dialect)
+2. Only mention properties from the Available Properties list above
+3. If no suitable properties exist, say "عذراً، ما في عندي شقة بهاي المواصفات حالياً"
+4. Be natural and conversational
+5. Recommend the best matching properties
+
+Response:"""
+    
+    return prompt
+
+def extract_entities_basic(query: str) -> Dict[str, Any]:
+    """Extract basic entities from Syrian Arabic query"""
+    entities = {
+        'location': None,
+        'furnished': None,
+        'rooms': None,
+        'features': [],
+        'price_pref': None
+    }
+    
+    query_lower = query.lower()
+    
+    # Location extraction
+    locations = ['المالكي', 'باب شرقي', 'المزة', 'الشعلان', 'أبو رمانة', 'القصاع', 'الجرمانا', 'باب توما']
+    for loc in locations:
+        if loc in query_lower:
+            entities['location'] = loc
+            break
+    
+    # Furnished status
+    if 'مفروش' in query_lower:
+        entities['furnished'] = 'furnished'
+    elif 'فارغ' in query_lower:
+        entities['furnished'] = 'unfurnished'
+    
+    # Room count
+    if 'ستوديو' in query_lower:
+        entities['rooms'] = 'studio'
+    elif 'غرفتين' in query_lower:
+        entities['rooms'] = '2_rooms'
+    elif 'ثلاث غرف' in query_lower:
+        entities['rooms'] = '3_rooms'
+    
+    # Features
+    if 'شرفة' in query_lower:
+        entities['features'].append('balcony')
+    if 'موقف' in query_lower:
+        entities['features'].append('parking')
+    if 'مكيف' in query_lower:
+        entities['features'].append('ac')
+    
+    # Price preference
+    if 'رخيص' in query_lower:
+        entities['price_pref'] = 'cheap'
+    elif 'فاخر' in query_lower:
+        entities['price_pref'] = 'luxury'
+    
+    return entities
+
+def filter_listings_by_entities(listings: List[str], entities: Dict[str, Any]) -> List[str]:
+    """Filter listings based on extracted entities"""
+    filtered = []
+    
+    for listing in listings:
+        listing_lower = listing.lower()
+        matches = True
+        
+        # Location filter
+        if entities.get('location'):
+            if entities['location'] not in listing_lower:
+                matches = False
+        
+        # Furnished filter
+        if entities.get('furnished'):
+            if entities['furnished'] == 'furnished' and 'مفروش' not in listing_lower:
+                matches = False
+            elif entities['furnished'] == 'unfurnished' and 'فارغ' not in listing_lower:
+                matches = False
+        
+        # Room filter
+        if entities.get('rooms'):
+            if entities['rooms'] == 'studio' and 'ستوديو' not in listing_lower:
+                matches = False
+            elif entities['rooms'] == '2_rooms' and 'غرفتين' not in listing_lower:
+                matches = False
+            elif entities['rooms'] == '3_rooms' and 'ثلاث غرف' not in listing_lower:
+                matches = False
+        
+        # Feature filters
+        for feature in entities.get('features', []):
+            if feature == 'balcony' and 'شرفة' not in listing_lower:
+                matches = False
+            elif feature == 'parking' and 'موقف' not in listing_lower:
+                matches = False
+            elif feature == 'ac' and 'مكيف' not in listing_lower:
+                matches = False
+        
+        if matches:
+            filtered.append(listing)
+    
+    return filtered[:5]  # Return top 5 matches
+
+def generate_response(user_query: str) -> tuple:
+    """Generate response using Falcon 7B"""
+    
+    if not model_loaded:
+        return "❌ Model not loaded. Please check the setup.", "{}", "[]"
+    
+    try:
+        # Extract entities
+        entities = extract_entities_basic(user_query)
+        
+        # Filter listings
+        relevant_listings = filter_listings_by_entities(DAMASCUS_LISTINGS, entities)
+        
+        # Create prompt
+        prompt = create_falcon_prompt(user_query, relevant_listings if relevant_listings else DAMASCUS_LISTINGS[:3])
+        
+        # Generate with Falcon
+        response = generator(
+            prompt,
+            max_length=len(prompt) + MAX_LENGTH,
+            temperature=TEMPERATURE,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id,
+            num_return_sequences=1
+        )
+        
+        # Extract generated text
+        full_response = response[0]['generated_text']
+        generated_part = full_response[len(prompt):].strip()
+        
+        # Clean up response
+        if generated_part:
+            # Remove any potential English or unwanted text
+            lines = generated_part.split('\n')
+            arabic_lines = []
+            for line in lines:
+                if line.strip() and any('\u0600' <= c <= '\u06FF' for c in line):
+                    arabic_lines.append(line.strip())
+                elif line.strip() and not any(c.isalpha() for c in line):
+                    arabic_lines.append(line.strip())  # Keep Arabic numerals/punctuation
+            
+            final_response = '\n'.join(arabic_lines[:5])  # Limit response length
+            
+            if not final_response:
+                final_response = "عذراً، صار في مشكلة بالنظام. جرب مرة تانية."
+        else:
+            final_response = "عذراً، ما قدرت أفهم طلبك. ممكن تعيد صياغته؟"
+        
+        return final_response, json.dumps(entities, ensure_ascii=False), json.dumps(relevant_listings, ensure_ascii=False)
+        
+    except Exception as e:
+        return f"❌ Error: {str(e)}", "{}", "[]"
+
+# Gradio interface
+def chatbot_interface(message: str) -> tuple:
+    """Gradio interface function"""
+    response, entities, listings = generate_response(message)
+    return response, entities, listings
+
+# Create Gradio app
+with gr.Blocks(title="Damascus Real Estate - Falcon 7B") as demo:
+    gr.Markdown("# 🏠 Damascus Real Estate Chatbot")
+    gr.Markdown("### Powered by Falcon 7B Instruct - Syrian Arabic Dialect")
+    gr.Markdown("Ask about apartments in Damascus in Arabic!")
+    
+    with gr.Row():
+        with gr.Column(scale=2):
+            user_input = gr.Textbox(
+                label="سؤالك بالعربي (Your Question in Arabic)",
+                placeholder="مثال: بدي شقة مفروشة بالمالكي مع شرفة",
+                lines=2
+            )
+            submit_btn = gr.Button("🔍 ابحث (Search)", variant="primary")
+            
+        with gr.Column(scale=3):
+            chatbot_response = gr.Textbook(
+                label="🤖 رد الوكيل العقاري (Real Estate Agent Response)",
+                lines=6
+            )
+    
+    with gr.Row():
+        with gr.Column():
+            extracted_entities = gr.Textbox(
+                label="📋 المتطلبات المستخرجة (Extracted Requirements)",
+                lines=3
+            )
+        with gr.Column():
+            matched_listings = gr.Textbox(
+                label="🏠 الشقق المطابقة (Matched Properties)",
+                lines=3
+            )
+    
+    # Example queries
+    gr.Markdown("### أمثلة (Examples):")
+    examples = [
+        "بدي شقة مفروشة بالمالكي مع شرفة واسعة",
+        "في شي ستوديو باب شرقي مفروش؟",
+        "بدي شقة ثلاث غرف فارغة بالمزة",
+        "شو عندك بالشعلان مفروش؟",
+        "بدي شقة رخيصة مع مكيف"
+    ]
+    
+    for example in examples:
+        gr.Button(example).click(
+            lambda x=example: x, 
+            outputs=user_input
+        )
+    
+    # Event handlers
+    submit_btn.click(
+        chatbot_interface,
+        inputs=user_input,
+        outputs=[chatbot_response, extracted_entities, matched_listings]
+    )
+    
+    user_input.submit(
+        chatbot_interface,
+        inputs=user_input,
+        outputs=[chatbot_response, extracted_entities, matched_listings]
+    )
+
+# Launch the app
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,8 @@
+torch
+torchvision
+torchaudio
+transformers==4.31.0
+gradio
+accelerate
+einops
+xformers