modified args

app.py

@@ -23,12 +23,12 @@ model.eval()
 pipe = pipeline('text-generation', model=model, tokenizer=tokenizer, eos_token_id=tokenizer.eos_token_id)
 
 def predict(text):
-  stopping_cond = StoppingCriteriaList([tokenizer.encode('<|endoftext|>')])
   with torch.no_grad():
     tokens = tokenizer(text, return_tensors="pt").input_ids
+    # generate and end generate if <|endoftext|> is not in text
     gen_tokens = model.generate(
       tokens, do_sample=True, temperature=0.8, max_new_tokens=64, top_k=50, top_p=0.8,
-      no_repeat_ngram_size=3, repetition_penalty=1.2  
+      no_repeat_ngram_size=3, repetition_penalty=1.2, bad_word_ids=[[11066]], eos_token_id=tokenizer.eos_token_id
     )
     generated = tokenizer.batch_decode(gen_tokens)[0]
     return generated