Create app.py

app.py

@@ -0,0 +1,215 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
+import torch
+import gc
+
+# =============================================================================
+# إعداد النموذج (يحمّل مرة واحدة فقط عند بدء الخادم)
+# =============================================================================
+
+MODEL_ID = "WeiboAI/VibeThinker-1.5B"
+
+print(f"✅ [إعداد]: جاري تحميل النموذج {MODEL_ID}...")
+print("    - هذا قد يستغرق 1-3 دقائق أول مرة...")
+
+try:
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        torch_dtype=torch.bfloat16,
+        device_map="auto",
+        low_cpu_mem_usage=True,
+        trust_remote_code=True
+    )
+    print("✅ [جاهز]: تم تحميل النموذج بنجاح!")
+except Exception as e:
+    print(f"❌ [خطأ]: فشل تحميل النموذج: {e}")
+    raise
+
+# =============================================================================
+# دوال التوليد
+# =============================================================================
+
+def generate_response(prompt, temperature=0.6, max_tokens=2048):
+    """توليد إجابة من النموذج"""
+    try:
+        messages = [{"role": "user", "content": prompt}]
+        text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+        inputs = tokenizer([text], return_tensors="pt").to(model.device)
+        
+        generation_config = {
+            "max_new_tokens": max_tokens,
+            "do_sample": True,
+            "temperature": temperature,
+            "top_p": 0.95,
+            "top_k": None
+        }
+        
+        with torch.no_grad():
+            outputs = model.generate(**inputs, generation_config=GenerationConfig(**generation_config))
+        
+        response = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
+        # استخراج الإجابة فقط (إزالة Prompt)
+        response = response.split("assistant\n")[-1].strip()
+        
+        return response
+    except Exception as e:
+        return f"❌ حدث خطأ: {str(e)}"
+
+def clear_memory():
+    """تنظيف الذاكرة"""
+    gc.collect()
+    torch.cuda.empty_cache()
+    return "✅ تم تنظيف الذاكرة"
+
+# =============================================================================
+# واجهة Gradio
+# =============================================================================
+
+css = """
+/* تنسيق مشابه للـ Colab */
+body { background: linear-gradient(135deg, #1a1a2e, #16213e); color: white; }
+.gradio-container { background: rgba(0,0,0,0.3); border-radius: 15px; }
+"""
+
+with gr.Blocks(title="VibeThinker API | HF Space", css=css, theme=gr.themes.Soft()) as demo:
+    
+    gr.Markdown("""
+    # 🚀 VibeThinker-1.5B API على Hugging Face Space
+    
+    **نموذج التفكير المنطقي والرياضيات من WeiboAI**
+    
+    ⚡ **النموذج**: `WeiboAI/VibeThinker-1.5B`
+    🔗 **المصدر**: [Hugging Face Hub](https://huggingface.co/WeiboAI/VibeThinker-1.5B)
+    📄 **الترخيص**: MIT (مجاني للاستخدام التجاري)
+    """)
+    
+    with gr.Row():
+        with gr.Column(scale=4):
+            chatbot = gr.Chatbot(
+                label="💬 محادثة",
+                height=500,
+                bubble_full_width=False,
+                avatar_images=(None, "https://huggingface.co/WeiboAI/VibeThinker-1.5B/resolve/main/logo.png")
+            )
+            
+            with gr.Row():
+                prompt_input = gr.Textbox(
+                    label="اكتب سؤالك هنا...",
+                    placeholder="أدخل مسألة رياضية أو سؤال منطقي...",
+                    lines=2,
+                    scale=8
+                )
+                submit_btn = gr.Button("🚀 إرسال", scale=1, variant="primary")
+            
+            with gr.Accordion("⚙️ إعدادات متقدمة", open=False):
+                temperature = gr.Slider(0.1, 1.5, value=0.6, label="Temperature", info="درجة الإبداع")
+                max_tokens = gr.Slider(512, 8192, value=2048, label="Max Tokens", info="أقصى طول للإجابة")
+                clear_btn = gr.Button("🗑️ مسح الذاكرة", variant="stop")
+        
+        with gr.Column(scale=1):
+            gr.Markdown("### 📋 معلومات")
+            gr.Markdown("""
+            - **سرعة الاستجابة**: ~2-5 ثانية/سؤال
+            - **دعم اللغة**: الإنجليزية (المهام الرياضية)
+            - **استخدام الذاكرة**: ~4GB VRAM
+            """)
+            
+            status_box = gr.Textbox(
+                label="حالة النظام",
+                value="✅ النموذج جاهز",
+                interactive=False
+            )
+    
+    # =============================================================================
+    # أحداث التحكم
+    # =============================================================================
+    
+    def chat_function(message, history, temp, max_tok):
+        """معالجة المحادثة"""
+        # بناء السياق من المحادثة السابقة
+        full_prompt = ""
+        for human, assistant in history:
+            full_prompt += f"Human: {human}\nAssistant: {assistant}\n"
+        full_prompt += f"Human: {message}\nAssistant: "
+        
+        response = generate_response(full_prompt, temp, max_tok)
+        return response
+    
+    def handle_submit(prompt, history, temp, max_tok):
+        """معالجة إرسال السؤال"""
+        if not prompt.strip():
+            return "", history
+        
+        # إضافة السؤال للمحادثة
+        history.append([prompt, None])
+        
+        # توليد الإجابة
+        response = chat_function(prompt, history[:-1], temp, max_tok)
+        
+        # تحديث المحادثة بالإجابة
+        history[-1][1] = response
+        
+        # تنظيف الذاكرة بعد كل استجابة
+        clear_memory()
+        
+        return "", history
+    
+    # ربط الأحداث
+    submit_btn.click(
+        handle_submit,
+        inputs=[prompt_input, chatbot, temperature, max_tokens],
+        outputs=[prompt_input, chatbot]
+    )
+    
+    prompt_input.submit(
+        handle_submit,
+        inputs=[prompt_input, chatbot, temperature, max_tokens],
+        outputs=[prompt_input, chatbot]
+    )
+    
+    clear_btn.click(
+        clear_memory,
+        outputs=[status_box]
+    )
+    
+    # =============================================================================
+    # معلومات مخصصة لـ API
+    # =============================================================================
+    
+    gr.Markdown("""
+    ---
+    ### 🔌 **استخدام كـ API**
+    
+    يمكنك استخدام هذا Space كـ API خارجي:
+    
+    **Endpoint**: `https://YOUR-SPACE-NAME.hf.space/v1/chat/completions`
+    
+    **Headers**:
+    ```json
+    {
+      "Content-Type": "application/json"
+    }
+    ```
+    
+    **Body**:
+    ```json
+    {
+      "model": "VibeThinker-1.5B",
+      "messages": [{"role": "user", "content": "مسألتك"}],
+      "temperature": 0.6
+    }
+    ```
+    """)
+
+# =============================================================================
+# تشغيل التطبيق
+# =============================================================================
+
+demo.queue(max_size=20).launch(
+    server_name="0.0.0.0",
+    server_port=7860,
+    share=False,  # لا تحتاج لـ share لأن Space عام
+    show_api=True,  # إظهار وثائق API تلقائياً
+    show_error=True
+)