Update app.py

app.py

@@ -22,8 +22,10 @@ model = AutoModelForCausalLM.from_pretrained("justinj92/phi3-orpo")
 
 terminators = [
     tok.eos_token_id,
-    '<|end|>',
-    '<|assistant|>'
+    32007,
+    32011,
+    32001,
+    32000
 ]
 
 if torch.cuda.is_available():
@@ -39,7 +41,7 @@ model = model.to(device)
 
 @spaces.GPU(duration=60)
 def chat(message, history, temperature, do_sample, max_tokens):
-    chat = []
+    chat = [{"role": "system", "content": "You are ORPO Tuned Phi Beast. Answer all questions in the most helpful way. No yapping."}]
     for item in history:
         chat.append({"role": "user", "content": item[0]})
         if item[1] is not None: