upadte

app.py

@@ -17,25 +17,23 @@ def load_system_prompt():
 
 DEFAULT_SYSTEM_PROMPT = load_system_prompt()
 
-# 🔄 تغيير المسار للموديل الجديد
 model_path = "unsloth/gemma-3-4b-it-unsloth-bnb-4bit"
 
 # إذا كان فيه HF_TOKEN في البيئة
 hf_token = os.getenv("HF_TOKEN")
 
-# ✅ Initialize pipeline for chat - إصلاح مشكلة torch_dtype
+# Initialize pipeline for chat
 pipeline_model = pipeline(
-    "image-text-to-text",
+    "text-generation",
     model=model_path,
-    device=0,  # Use GPU device directly
+    device_map="auto",
     token=hf_token,
-    trust_remote_code=True,
-  
+    trust_remote_code=True
 )
 
 def generate_with_pipeline(messages, max_new_tokens=256, temperature=0.7, top_p=0.9, top_k=50, repetition_penalty=1.0):
     """Generate response using the pipeline with messages format"""
-    # Apply chat template (Gemma uses its own template)
+    # Apply chat template
     prompt = pipeline_model.tokenizer.apply_chat_template(
         messages,
         tokenize=False,
@@ -50,8 +48,7 @@ def generate_with_pipeline(messages, max_new_tokens=256, temperature=0.7, top_p=
         top_k=top_k,
         repetition_penalty=repetition_penalty,
         do_sample=True,
-        return_full_text=False,
-        eos_token_id=pipeline_model.tokenizer.eos_token_id,
+        return_full_text=False
     )
     return outputs[0]["generated_text"]
 
@@ -67,31 +64,15 @@ def generate_response(message, history, max_new_tokens, temperature, top_p, top_
         max_new_tokens, temperature, top_p, top_k, repetition_penalty: Generation parameters
     """
     try:
-        # ✅ Build messages list with system prompt as first user message
-        messages = []
-        
-        # ✅ System prompt as first user message + model acknowledgment
-        messages.append({
-            "role": "user",
-            "content": DEFAULT_SYSTEM_PROMPT
-        })
-        messages.append({
-            "role": "model",  # ✅ في Gemma 3 استخدم "model" مو "assistant"
-            "content": "Understood. I will follow these instructions."
-        })
+        # Build messages list starting with system prompt
+        messages = [{"role": "system", "content": DEFAULT_SYSTEM_PROMPT}]
         
         # Add conversation history
+        # When type="messages", history is a list of message dicts with 'role' and 'content'
         if history:
             for msg in history:
                 if isinstance(msg, dict) and 'role' in msg and 'content' in msg:
-                    # ✅ تحويل assistant → model إذا لزم
-                    role = msg['role']
-                    if role == 'assistant':
-                        role = 'model'
-                    messages.append({
-                        "role": role,
-                        "content": msg['content']
-                    })
+                    messages.append({"role": msg['role'], "content": msg['content']})
         
         # Add current user message
         if isinstance(message, dict):
@@ -99,10 +80,7 @@ def generate_response(message, history, max_new_tokens, temperature, top_p, top_
         else:
             current_message = str(message)
         
-        messages.append({
-            "role": "user",
-            "content": current_message
-        })
+        messages.append({"role": "user", "content": current_message})
         
         # Debug: print messages structure
         print(f"Messages sent to model: {len(messages)} messages")
@@ -134,9 +112,9 @@ demo = gr.ChatInterface(
     fn=generate_response,
     additional_inputs=[
         gr.Slider(label="الحد الأقصى للكلمات الجديدة", minimum=64, maximum=4096, step=1, value=2048),
-        gr.Slider(label="درجة الحرارة", minimum=0.1, maximum=2.0, step=0.1, value=1.0),
-        gr.Slider(label="Top-p", minimum=0.05, maximum=1.0, step=0.05, value=0.95),
-        gr.Slider(label="Top-k", minimum=1, maximum=100, step=1, value=64),
+        gr.Slider(label="درجة الحرارة", minimum=0.1, maximum=2.0, step=0.1, value=0.7),
+        gr.Slider(label="Top-p", minimum=0.05, maximum=1.0, step=0.05, value=0.9),
+        gr.Slider(label="Top-k", minimum=1, maximum=100, step=1, value=50),
         gr.Slider(label="عقوبة التكرار", minimum=1.0, maximum=2.0, step=0.05, value=1.0)
     ],
     examples=[
@@ -148,7 +126,7 @@ demo = gr.ChatInterface(
     ],
     cache_examples=False,
     type="messages",
-    title="دعم عملاء TechSolutions - مساعد أليكس (Google Gemma 3)",
+    title="دعم عملاء TechSolutions - مساعد أليكس (العراقي)",
     description="""🤖 مساعد خدمة عملاء ذكي لـ TechSolutions
 
 ✨ المميزات:
@@ -157,8 +135,7 @@ demo = gr.ChatInterface(
 - 🔧 دعم فني واستكشاف الأخطاء
 - 📋 معلومات الخدمات والإرشاد
 - 🧠 **يتذكر المحادثة السابقة** - يمكنك الرجوع للمواضيع السابقة
-- 🎯 مدعوم بـ **Google Gemma 3 4B** (أسرع وأخف من Llama!)
-- ⚡ تحسينات Unsloth للأداء والضغط
+- 🎯 مدعوم بـ موديل Unsloth Meta-Llama-3.1-8B-Instruct-bnb-4bit
 
 احجي مع أليكس لحل مشاكلك التقنية، استفسر عن الخدمات، أو احصل على معلومات المنتجات.""",
     fill_height=True,