Update app.py

app.py

@@ -9,7 +9,7 @@ model = AutoModelForCausalLM.from_pretrained("KoboldAI/OPT-2.7B-Nerybus-Mix")
 def chat(Prompt):
 
     input_ids = tokenizer(Prompt, return_tensors="pt").input_ids
-    generated_ids = model.generate(input_ids, max_length=256, repetition_penalty = 2.5, top_k = 75, eos_token_id=tokenizer.eos_token_id)
+    generated_ids = model.generate(input_ids, use_cache=True, max_length=256, repetition_penalty = 2.5, top_k = 75, eos_token_id=tokenizer.eos_token_id)
     bot = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
     return bot