app.py

app.py

@@ -0,0 +1,55 @@
+import gradio as gr
+from transformers import PreTrainedTokenizerFast, BartForConditionalGeneration
+
+# 1. 모델 & 토크나이저 로드
+MODEL_NAME = "gangyeolkim/kobart-korean-summarizer-v2"
+tokenizer = PreTrainedTokenizerFast.from_pretrained(MODEL_NAME)
+model = BartForConditionalGeneration.from_pretrained(MODEL_NAME)
+
+# 2. 요약 함수
+def summarize(text, min_len, max_len):
+    if not text.strip():
+        return "⚠️ 요약할 텍스트를 입력하세요."
+    
+    # 토큰화
+    inputs = tokenizer(
+        [text],
+        max_length=1024,
+        truncation=True,
+        return_tensors="pt"
+    )
+
+    # 모델 추론
+    summary_ids = model.generate(
+        inputs["input_ids"],
+        num_beams=4,
+        min_length=min_len,
+        max_length=max_len,
+        early_stopping=True
+    )
+
+    # 디코딩
+    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
+    return summary
+
+# 3. Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("## 📝 KoBART 한국어 요약기 (CPU 최적화 가능)")
+    with gr.Row():
+        with gr.Column():
+            input_text = gr.Textbox(
+                label="원문 입력 (최대 2000자)",
+                lines=15,
+                placeholder="여기에 요약할 한국어 텍스트를 입력하세요."
+            )
+            min_len = gr.Slider(50, 500, value=100, step=10, label="최소 요약 길이")
+            max_len = gr.Slider(100, 1500, value=300, step=10, label="최대 요약 길이")
+            btn = gr.Button("요약 실행")
+        with gr.Column():
+            output_text = gr.Textbox(label="요약 결과", lines=15)
+
+    btn.click(summarize, inputs=[input_text, min_len, max_len], outputs=output_text)
+
+# 4. 실행
+if __name__ == "__main__":
+    demo.launch()