Update app.py

app.py

@@ -1,43 +1,65 @@
+import os
 import gradio as gr
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+from huggingface_hub import snapshot_download
 
-# ------------------------------
-# Configuration
-# ------------------------------
-MODEL_NAME = "tiiuae/Falcon-H1-7B-Instruct"
-MAX_LENGTH = 120       # safer than 50
+# ================================
+# CONFIGURATION
+# ================================
+MODEL_NAME_PRIMARY = "tiiuae/Falcon-H1-7B-Instruct"
+MODEL_NAME_FALLBACK = "tiiuae/falcon-7b-instruct"
+MODEL_LOCAL_DIR = "./falcon_model"
+MAX_LENGTH = 120
 TEMPERATURE = 0.3
 REPETITION_PENALTY = 1.8
 
-print("🚀 Loading Falcon 7B Arabic Instruct...")
+print("🚀 Preparing environment...")
 
-# ------------------------------
-# Load model and tokenizer
-# ------------------------------
+# 1️⃣ Upgrade transformers & accelerate
+os.system("pip install --upgrade transformers accelerate > /dev/null")
+
+# 2️⃣ Ensure clean download of model
 try:
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+    print(f"⬇️ Downloading model: {MODEL_NAME_PRIMARY}")
+    snapshot_download(MODEL_NAME_PRIMARY, local_dir=MODEL_LOCAL_DIR, force_download=True)
+    model_name = MODEL_LOCAL_DIR
+except Exception as e:
+    print(f"⚠️ Primary model download failed: {e}")
+    print("➡️ Falling back to Falcon 7B Instruct")
+    snapshot_download(MODEL_NAME_FALLBACK, local_dir=MODEL_LOCAL_DIR, force_download=True)
+    model_name = MODEL_LOCAL_DIR
+
+# 3️⃣ Load tokenizer and model
+try:
+    print("🔄 Loading tokenizer and model...")
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
     model = AutoModelForCausalLM.from_pretrained(
-        MODEL_NAME,
+        model_name,
         torch_dtype=torch.float16,
+        trust_remote_code=True,
         device_map="auto",
-        trust_remote_code=True
+        low_cpu_mem_usage=True
     )
-    
-    print("✅ Falcon 7B Arabic model loaded successfully")
-    model_loaded = True
 
-    print("✅ Falcon 7B Arabic model loaded successfully")
+    generator = pipeline(
+        "text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        torch_dtype=torch.float16,
+        device=0 if torch.cuda.is_available() else -1
+    )
+    print("✅ Model loaded successfully")
     model_loaded = True
-    
+
 except Exception as e:
     print(f"❌ Model loading failed: {e}")
     generator = None
     model_loaded = False
 
-# ------------------------------
+# ================================
 # Test Questions (Pre-Filled)
-# ------------------------------
+# ================================
 test_questions = [
     "بدي شقة بالمالكي فيها شرفة وغسالة صحون.",
     "هل في شقة دوبلكس بالمزة الفيلات فيها موقفين سيارة؟",
@@ -48,14 +70,13 @@ test_questions = [
     "عندك منزل مستقل بالمهاجرين مع موقد حطب؟"
 ]
 
-# ------------------------------
+# ================================
 # Falcon Chat Function
-# ------------------------------
+# ================================
 def chat_falcon(user_input):
     if not model_loaded:
         return "❌ النموذج غير محمل. تحقق من الإعدادات."
-    
-    # Structured prompt for clarity
+
     prompt = f"أنت مساعد عقارات ذكي. أجب بجملة أو جملتين واضحتين.\nالسؤال: {user_input}\nالجواب:"
     
     output = generator(
@@ -66,16 +87,14 @@ def chat_falcon(user_input):
         repetition_penalty=REPETITION_PENALTY,
         top_p=0.9
     )[0]["generated_text"]
-    
-    # Strip the prompt part from the output
-    cleaned_output = output.replace(prompt, "").strip()
-    return cleaned_output
 
-# ------------------------------
+    return output.replace(prompt, "").strip()
+
+# ================================
 # Build Gradio Interface
-# ------------------------------
+# ================================
 with gr.Blocks() as demo:
-    gr.Markdown("## 🏠 Falcon 7B Arabic Instruct - Damascus Real Estate Test")
+    gr.Markdown("## 🏠 Falcon H1 7B Instruct - Damascus Real Estate Test")
     gr.Markdown("اختبر قدرة النموذج على فهم الأسئلة بالعربية (لهجة سورية أو فصحى)")
 
     with gr.Row():
@@ -84,10 +103,10 @@ with gr.Blocks() as demo:
             submit_btn = gr.Button("🔎 أرسل")
         with gr.Column(scale=1):
             suggestions = gr.Dropdown(choices=test_questions, label="🧾 أسئلة جاهزة", value=test_questions[0])
-    
+
     output_box = gr.Textbox(label="إجابة النموذج", lines=8)
-    
+
     submit_btn.click(fn=chat_falcon, inputs=user_input, outputs=output_box)
     suggestions.change(fn=chat_falcon, inputs=suggestions, outputs=output_box)
 
-demo.launch(share=True)
+demo.launch()