upadte

app.py

@@ -23,17 +23,15 @@ 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
-# ✅ استخدم image-text-to-text حتى لو text-only
+# ✅ Initialize pipeline for chat - إصلاح مشكلة torch_dtype
 pipeline_model = pipeline(
-    "image-text-to-text",  # ✅ الصحيح لـ Gemma 3
+    "image-text-to-text",
     model=model_path,
     device=0,  # Use GPU device directly
-    torch_dtype=torch.bfloat16,
     token=hf_token,
     trust_remote_code=True,
     model_kwargs={
-        "torch_dtype": torch.bfloat16,
+        "torch_dtype": torch.bfloat16,  # ✅ فقط هنا
         "load_in_4bit": True,
         "bnb_4bit_compute_dtype": torch.bfloat16,
         "bnb_4bit_use_double_quant": False,
@@ -114,8 +112,6 @@ def generate_response(message, history, max_new_tokens, temperature, top_p, top_
         
         # Debug: print messages structure
         print(f"Messages sent to model: {len(messages)} messages")
-        for i, msg in enumerate(messages):
-            print(f"  {i}: {msg['role']}: {msg['content'][:50]}...")
         
         # Generate response
         response = generate_with_pipeline(
@@ -144,10 +140,10 @@ 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),  # ✅ Gemma يفضل 1.0
+        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),  # ✅ Gemma يفضل 64
-        gr.Slider(label="عقوبة التكرار", minimum=1.0, maximum=2.0, step=0.05, value=1.0)  # ✅ 1.0 = disabled
+        gr.Slider(label="Top-k", minimum=1, maximum=100, step=1, value=64),
+        gr.Slider(label="عقوبة التكرار", minimum=1.0, maximum=2.0, step=0.05, value=1.0)
     ],
     examples=[
         ["النت عندي معطل من الصبح، تقدر تساعدني؟"],