duration

app.py

@@ -14,40 +14,18 @@ model = AutoModelForCausalLM.from_pretrained(
     trust_remote_code=True,
 )
 
-@spaces.GPU  # 이 함수 실행될 때 GPU 할당됨!
-def chat(user_input):
-    model.to("cuda")  # 함수 안에서 GPU로 이동!
-    
-    messages = [{
-        "role": "user",
-        "content": [{"type": "text", "text": user_input}]
-    }]
-    prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
+@spaces.GPU(duration=300)
+def generate_response(prompt):
+    # 모델 및 토크나이저 로딩은 함수 내부에서 수행
+    tokenizer = AutoTokenizer.from_pretrained("dasomaru/gemma-3-4bit-it-demo")
+    model = AutoModelForCausalLM.from_pretrained("dasomaru/gemma-3-4bit-it-demo")
+    model.to("cuda")
 
     inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+    outputs = model.generate(**inputs, max_new_tokens=64)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=256,
-            temperature=1.0,
-            top_p=0.95,
-            top_k=64,
-            do_sample=True,
-        )
-
-    output_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    return output_text.split(user_input)[-1].strip()
-
-# Gradio 인터페이스 설정
-demo = gr.Interface(
-    fn=chat,
-    inputs=gr.Textbox(lines=2, placeholder="Type your message..."),
-    outputs=gr.Textbox(lines=10),
-    title="🧠 Gemma-3 4bit (ZeroGPU)",
-    description="This Space uses the ZeroGPU feature. First request might take a few seconds!"
-)
-
+demo = gr.Interface(fn=generate_response, inputs="text", outputs="text")
 demo.launch()