2

app.py

@@ -2,7 +2,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 import torch
 import gradio as gr
 
-model_name = "beomi/KoAlpaca-Polyglot-5.8B"
+model_name = "beomi/KoAlpaca-7B"
 
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(
@@ -15,18 +15,21 @@ pipe = pipeline(
     "text-generation",
     model=model,
     tokenizer=tokenizer,
-    max_new_tokens=256,
+    max_new_tokens=128,
     do_sample=True,
-    temperature=0.7,
+    temperature=0.5,
     top_p=0.9,
 )
 
 def chat_fn(prompt):
     try:
-        output = pipe(prompt)[0]["generated_text"]
-        return output
+        outputs = pipe(prompt)
+        # text-generation 파이프라인의 출력은 리스트 형태로 반환되며,
+        # 각 출력은 generated_text 키를 포함합니다
+        return outputs[0]["generated_text"]
     except Exception as e:
-        return [f"Error: {str(e)}"]
+        print(f"Error in chat_fn: {str(e)}")  # 디버깅을 위한 에러 로깅 추가
+        return f"Error: {str(e)}"
 
 
 with gr.Blocks() as demo:
@@ -47,7 +50,12 @@ with gr.Blocks() as demo:
     demo.load(chat_fn, inputs=input_box, outputs=output_box)
 
 # ✅ API endpoint로 사용할 Interface 객체 등록
-api_demo = gr.Interface(fn=chat_fn, inputs="text", outputs="text", name="predict")
+api_demo = gr.Interface(
+    fn=chat_fn,
+    inputs="text",
+    outputs="text",
+    api_name="predict"  # API 엔드포인트 이름 명시
+)
 
 if __name__ == "__main__":
     demo.queue()