Update app.py

app.py

@@ -1,10 +1,55 @@
 import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-with gr.Blocks(fill_height=True) as demo:
-    with gr.Sidebar():
+model_name = "kakaocorp/kanana-nano-2.1b-instruct"
+
+# 모델과 토크나이저를 CPU 환경에서 로드합니다.
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float32,  # CPU에서는 bfloat16 지원이 제한될 수 있으므로 float32 사용 권장
+    trust_remote_code=True,
+)
+# CPU만 사용하므로 .to("cuda") 부분은 생략합니다.
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+def generate_response(prompt):
+    messages = [
+        {"role": "system", "content": "You are a helpful AI assistant developed by Kakao."},
+        {"role": "user", "content": prompt}
+    ]
+    input_ids = tokenizer.apply_chat_template(
+        messages,
+        tokenize=True,
+        add_generation_prompt=True,
+        return_tensors="pt"
+    )
+    
+    model.eval()
+    with torch.no_grad():
+        output = model.generate(
+            input_ids,
+            max_new_tokens=72,
+            do_sample=False,
+        )
+    return tokenizer.decode(output[0], skip_special_tokens=True)
+
+with gr.Blocks() as demo:
+    with gr.Tab("About"):
         gr.Markdown("# Inference Provider")
-        gr.Markdown("This Space showcases the kakaocorp/kanana-nano-2.1b-instruct model, served by the hf-inference API. Sign in with your Hugging Face account to use this API.")
-        button = gr.LoginButton("Sign in")
-    gr.load("models/kakaocorp/kanana-nano-2.1b-instruct", accept_token=button, provider="hf-inference")
+        gr.Markdown("이 Space는 kakaocorp/kanana-nano-2.1b-instruct 모델을 CPU에서 추론합니다.")
     
-demo.launch()
+    with gr.Tab("Generate"):
+        prompt_input = gr.Textbox(
+            label="Prompt 입력",
+            placeholder="여기에 프롬프트를 입력하세요...",
+            lines=5
+        )
+        generate_btn = gr.Button("생성")
+        output_text = gr.Textbox(
+            label="모델 출력",
+            lines=10
+        )
+        generate_btn.click(fn=generate_response, inputs=prompt_input, outputs=output_text)
+
+demo.launch()