Revert app.py to non ZeroGPU

app.py

@@ -2,33 +2,23 @@ import gradio as gr
 from transformers import GPT2LMHeadModel, AutoTokenizer, pipeline
 import torch
 
-def create_model():
-    # Initialize model with device-agnostic loading
-    model = GPT2LMHeadModel.from_pretrained(
-        "./model/gpt2-355M",
-        device_map="auto",  # Enables automatic device mapping
-        torch_dtype=torch.float16  # Use float16 for better memory efficiency
-    )
-    
-    tokenizer = AutoTokenizer.from_pretrained(
-        "gpt2",
-        pad_token='<|endoftext|>'
-    )
-    
-    return pipeline(
-        "text-generation",
-        model=model,
-        tokenizer=tokenizer,
-        device_map="auto",
-        config={"max_length": 140}
-    )
-
-# Initialize the pipeline
-trump = create_model()
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+
+# load pretrained + finetuned GPT2
+model = GPT2LMHeadModel.from_pretrained("./model/gpt2-355M")
+model = model.to(device)
+
+# create tokenizer
+tokenizer = AutoTokenizer.from_pretrained(
+    "gpt2", 
+    pad_token='<|endoftext|>'
+)
+
+trump = pipeline("text-generation", model=model, tokenizer=tokenizer, config={"max_length":140})
 
 def generate(text):
     result = trump(text, num_return_sequences=1)
-    return result[0]["generated_text"].replace('"', '')
+    return result[0]["generated_text"].replace('"', '')  # remove quotation marks
 
 examples = [
     ["Why does the lying news media"],
@@ -43,5 +33,4 @@ demo = gr.Interface(
     examples=examples
 )
 
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()