Update app.py

app.py

@@ -1,45 +1,208 @@
-from llama_cpp import Llama
 import gradio as gr
+from llama_cpp import Llama
+import os
+
+# ============================================================
+# تحميل النموذج
+# ============================================================
+MODEL_REPO = "DavidAU/Qwen2.5-MOE-2X1.5B-DeepSeek-Uncensored-Censored-4B-gguf"
+MODEL_FILE = "Qwen2.5-MOE-2X1.5B-DeepSeek-Uncensored-Censored-4B-Q4_K_M.gguf"
+
+print("⏳ جاري تحميل النموذج...")
+
+llm = Llama.from_pretrained(
+    repo_id=MODEL_REPO,
+    filename=MODEL_FILE,
+    n_ctx=8192,        # حجم السياق
+    n_threads=4,       # عدد خيوط المعالجة
+    n_gpu_layers=-1,   # استخدام GPU إذا كان متوفراً (-1 = كل الطبقات)
+    verbose=False,
+)
+
+print("✅ تم تحميل النموذج بنجاح!")
+
+# ============================================================
+# دالة التوليد
+# ============================================================
+def chat(
+    message,
+    history,
+    system_prompt,
+    max_tokens,
+    temperature,
+    top_p,
+    repeat_penalty,
+):
+    # بناء المحادثة بصيغة ChatML
+    messages = [{"role": "system", "content": system_prompt}]
+    
+    for user_msg, assistant_msg in history:
+        messages.append({"role": "user", "content": user_msg})
+        if assistant_msg:
+            messages.append({"role": "assistant", "content": assistant_msg})
+    
+    messages.append({"role": "user", "content": message})
+
+    # توليد الرد
+    response = llm.create_chat_completion(
+        messages=messages,
+        max_tokens=int(max_tokens),
+        temperature=float(temperature),
+        top_p=float(top_p),
+        repeat_penalty=float(repeat_penalty),
+        stream=True,
+    )
+
+    partial = ""
+    for chunk in response:
+        delta = chunk["choices"][0]["delta"].get("content", "")
+        partial += delta
+        yield partial
+
+
+# ============================================================
+# واجهة Gradio
+# ============================================================
+with gr.Blocks(
+    title="Qwen2.5 MOE 4B - DeepSeek",
+    theme=gr.themes.Soft(primary_hue="violet"),
+    css="""
+    .gradio-container { max-width: 900px !important; margin: auto; }
+    .title-box { text-align: center; padding: 20px; }
+    footer { display: none !important; }
+    """,
+) as demo:
+
+    gr.HTML("""
+    <div class='title-box'>
+        <h1>🧠 Qwen2.5 MOE 2×1.5B DeepSeek</h1>
+        <p style='color: #666; font-size: 14px;'>
+            نموذج مزيج من الخبراء — نموذجان 1.5B يعملان معاً كنموذج 4B
+        </p>
+        <p style='color: #888; font-size: 12px;'>
+            DavidAU/Qwen2.5-MOE-2X1.5B-DeepSeek-Uncensored-Censored-4B
+        </p>
+    </div>
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=3):
+            chatbot = gr.Chatbot(
+                label="المحادثة",
+                height=500,
+                show_label=True,
+                bubble_full_width=False,
+            )
+            with gr.Row():
+                msg = gr.Textbox(
+                    placeholder="اكتب رسالتك هنا...",
+                    label="رسالتك",
+                    scale=4,
+                    lines=2,
+                )
+                send_btn = gr.Button("إرسال 🚀", scale=1, variant="primary")
+
+            clear_btn = gr.Button("مسح المحادثة 🗑️", variant="secondary")
+
+        with gr.Column(scale=1):
+            gr.Markdown("### ⚙️ الإعدادات")
+
+            system_prompt = gr.Textbox(
+                value="You are a helpful, creative, and intelligent assistant. Think carefully before answering.",
+                label="System Prompt",
+                lines=4,
+            )
+            max_tokens = gr.Slider(
+                minimum=64,
+                maximum=4096,
+                value=1024,
+                step=64,
+                label="Max Tokens (أقصى طول للرد)",
+            )
+            temperature = gr.Slider(
+                minimum=0.1,
+                maximum=2.0,
+                value=0.7,
+                step=0.05,
+                label="Temperature (الإبداع)",
+                info="0.4-0.8 للتفكير المنطقي | 1.0+ للإبداع",
+            )
+            top_p = gr.Slider(
+                minimum=0.1,
+                maximum=1.0,
+                value=0.95,
+                step=0.05,
+                label="Top P",
+            )
+            repeat_penalty = gr.Slider(
+                minimum=1.0,
+                maximum=1.5,
+                value=1.06,
+                step=0.01,
+                label="Repeat Penalty (تجنب التكرار)",
+            )
+
+            gr.Markdown("""
+            ### 💡 نصائح
+            - 🔵 **Temperature 0.4-0.8** → للتفكير والمنطق
+            - 🟣 **Temperature 1.0+** → للإبداع والقصص
+            - 📝 كن تفصيلياً في أسئلتك
+            - 🔄 جرب 2-4 مرات للحصول على أفضل نتيجة
+            """)
+
+    # ============================================================
+    # منطق الأحداث
+    # ============================================================
+    def user_submit(user_message, history):
+        return "", history + [[user_message, None]]
+
+    def bot_respond(history, system_prompt, max_tokens, temperature, top_p, repeat_penalty):
+        if not history or history[-1][1] is not None:
+            return history
+        
+        user_message = history[-1][0]
+        history_without_last = history[:-1]
+        
+        for partial_response in chat(
+            user_message,
+            history_without_last,
+            system_prompt,
+            max_tokens,
+            temperature,
+            top_p,
+            repeat_penalty,
+        ):
+            history[-1][1] = partial_response
+            yield history
+
+    msg.submit(
+        user_submit,
+        [msg, chatbot],
+        [msg, chatbot],
+    ).then(
+        bot_respond,
+        [chatbot, system_prompt, max_tokens, temperature, top_p, repeat_penalty],
+        chatbot,
+    )
 
-# 🔁 Lazy loading (important to avoid OOM)
-llm = None
-
-def load_model():
-    global llm
-    if llm is None:
-        llm = Llama.from_pretrained(
-            repo_id="bartowski/microsoft_Phi-4-mini-instruct-GGUF",
-            filename="microsoft_Phi-4-mini-instruct-Q4_K_M.gguf",
-            n_ctx=2048,      # lower = less RAM
-            n_threads=2,
-            verbose=False
-        )
-    return llm
-
-
-def chat(message, history):
-    model = load_model()
-
-    prompt = f"""<|im_start|>user
-{message}<|im_end|>
-<|im_start|>assistant
-"""
-
-    output = model(
-        prompt,
-        max_tokens=512,
-        temperature=0.7,
-        stop=["<|im_end|>"],
-        stream=True
+    send_btn.click(
+        user_submit,
+        [msg, chatbot],
+        [msg, chatbot],
+    ).then(
+        bot_respond,
+        [chatbot, system_prompt, max_tokens, temperature, top_p, repeat_penalty],
+        chatbot,
     )
 
-    response = ""
-    for chunk in output:
-        response += chunk["choices"][0]["text"]
-        yield response
+    clear_btn.click(lambda: [], None, chatbot)
 
+    gr.Markdown("""
+    ---
+    <div style='text-align:center; color: #aaa; font-size: 12px;'>
+    النموذج: <b>DavidAU/Qwen2.5-MOE-2X1.5B-DeepSeek-Uncensored-Censored-4B</b> | 
+    الصيغة: GGUF Q4_K_M | السياق: 8K
+    </div>
+    """)
 
-gr.ChatInterface(
-    chat,
-    title="🧠 Phi-4 Mini (CPU Stable Space)"
-).launch()
+demo.launch()