Update app.py

app.py

@@ -85,12 +85,13 @@ def chat(message, history):
     generation_kwargs = {
         "input_ids": input_ids,
         "streamer": streamer,
-        "max_new_tokens": 512,
-        "temperature": 0.7,
+        "max_new_tokens": 1024,
+        "temperature": 0.85,
         "top_p": 0.9,
         "top_k": 50,
         "do_sample": True,
         "repetition_penalty": 1.1,
+        "eos_token_id": None,  # ⬅️ مهم حتى لا يتوقف مبكراً
     }
 
     # ======================================================
@@ -102,6 +103,7 @@ def chat(message, history):
     partial_text = ""
     for new_text in streamer:
         partial_text += new_text
+        print(new_text, end="", flush=True)
         yield partial_text
 
     thread.join()