update

app.py

@@ -108,34 +108,49 @@ def generate_response(input_data, chat_history, max_new_tokens, temperature, top
     # Add current user message
     messages.append({"role": "user", "content": input_data})
 
-    # Validation للقيم عشان ما تحصل CUDA errors
-    temperature = max(0.1, min(2.0, temperature))  # 0.1 to 2.0
-    top_p = max(0.1, min(1.0, top_p))  # 0.1 to 1.0
-    top_k = max(1, min(100, top_k))  # 1 to 100
-    repetition_penalty = max(1.0, min(1.5, repetition_penalty))  # 1.0 to 1.5
-
-    # استخدام ChatPipeline المخصص مع streaming
-    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
-
-    generation_kwargs = pipe(
-        messages,
-        streamer=streamer,
-        max_new_tokens=max_new_tokens,
-        temperature=temperature,
-        top_p=top_p,
-        top_k=top_k,
-        repetition_penalty=repetition_penalty
-    )
-
-    thread = Thread(target=model.generate, kwargs=generation_kwargs)
-    thread.start()
-
-    # Stream the response
-    response = ""
-    for chunk in streamer:
-        response += chunk
+    # استخدام generate مباشرة مع parameters أكثر أماناً
+    try:
+        # محاولة استخدام chat template
+        if hasattr(tokenizer, 'apply_chat_template') and tokenizer.chat_template is not None:
+            prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+        else:
+            # Fallback format
+            prompt = f"System: {DEFAULT_SYSTEM_PROMPT}\n\n"
+            for msg in messages[1:]:  # Skip system message since we added it above
+                if msg["role"] == "user":
+                    prompt += f"Human: {msg['content']}\n"
+                elif msg["role"] == "assistant":
+                    prompt += f"Assistant: {msg['content']}\n"
+            prompt += "Assistant:"
+
+        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+
+        # استخدام generate مع parameters أساسية وآمنة
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=min(max_new_tokens, 512),  # حد أقصى أمان
+                do_sample=False,  # تعطيل sampling للأمان
+                num_beams=1,  # greedy decoding
+                pad_token_id=tokenizer.eos_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                return_dict_in_generate=True,
+                output_scores=False,
+            )
+
+        response = tokenizer.decode(outputs.sequences[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
+        response = response.strip()
+
+        if not response:
+            response = "آسف، حدث خطأ في توليد الرد. حاول مرة ثانية."
+
         yield response
 
+    except Exception as e:
+        error_msg = f"خطأ في التوليد: {str(e)}"
+        print(error_msg)
+        yield "آسف، حدث خطأ تقني. حاول مرة ثانية."
+
 demo = gr.ChatInterface(
     fn=generate_response,
     additional_inputs=[