modified some arguments

app.py

@@ -1,6 +1,7 @@
 import gradio as gr
-from transformers import pipeline
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM, StoppingCriteriaList
 import os
+import torch
 
 theme = "darkgrass"
 title = "GPT-NeoX(Korean) Demo"
@@ -14,10 +15,25 @@ examples = [
     ["2040년 미국은, "]
 ]
 
-pipe = pipeline('text-generation', model=model_name, use_auth_token=os.environ['TOKEN'])
+tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-ko-1.3b", use_auth_token=os.environ['TOKEN'])
+model = AutoModelForCausalLM.from_pretrained(
+  "EleutherAI/gpt-neox-ko-1.3b", use_auth_token=os.environ['TOKEN'],
+  pad_token_id=tokenizer.pad_token_id,
+  eos_token_id=tokenizer.eos_token_id,
+  bos_token_id=tokenizer.bos_token_id  
+)
+model.eval()
+
+pipe = pipeline('text-generation', model=model, tokenizer=tokenizer)
 
 def predict(text):
-  return pipe(text, do_sample=True, return_full_text=False, max_length=100)[0]['generated_text']
+  stopping_cond = StoppingCriteriaList([tokenizer.encode('<|endoftext|>')])
+  with torch.no_grad():
+    tokens = tokenizer(text, return_tensors="pt").input_ids
+    gen_tokens = model.generate(tokens, do_sample=True, temperature=0.8, max_new_tokens=64)
+    generated = tokenizer.batch_decode(gen_tokens)[0]
+    return generated
+  # return pipe(text)[0]['generated_text']
 
 iface = gr.Interface(
   fn=predict,