Update app.py

app.py

@@ -19,7 +19,8 @@ model = AutoModelForCausalLM.from_pretrained(
     model_name,
     dtype=torch.bfloat16,
     device_map="auto",
-    trust_remote_code=True
+    trust_remote_code=True,
+    attn_implementation="eager"  # تجنب مشاكل flash-attention
 )
 
 pipe = pipeline(
@@ -157,7 +158,11 @@ Explain your reasoning step-by-step in a clear forensic report.
     ]
     
     prompt = tokenizer.apply_chat_template(messages, tokenize=False)
-    response = pipe(prompt)[0]['generated_text']
+    
+    # ========== هنا الإصلاح الجذري: إضافة use_cache=False ==========
+    response = pipe(prompt, use_cache=False)[0]['generated_text']
+    # ================================================================
+    
     agent_answer = response.split("assistant")[-1].strip()
     
     discord_match = re.search(r'discord\.(gg|com/invite)/\S+', agent_answer, re.IGNORECASE)