Update app.py

app.py

@@ -15,7 +15,7 @@ import uvicorn
 import gradio as gr
 from threading import Thread
 from fastapi.middleware.cors import CORSMiddleware
-from transformers import AutoTokenizer, AutoModelForCausalLM
+
 #####################################
 # 1) 앱 및 모델 초기화
 #####################################
@@ -1114,36 +1114,50 @@ def recommend_content_based(user_profile: dict, top_n=5):
 #####################################
 # 5) 챗봇 로직
 #####################################
-tokenizer = AutoTokenizer.from_pretrained("Chanjeans/tfchatbot_2")
-model = AutoModelForCausalLM.from_pretrained("Chanjeans/tfchatbot_2")
-model.eval()
-print("Model loaded successfully.")
+HF_API_KEY = os.environ.get("HF_API_KEY", "YOUR_HF_API_KEY")
+API_URL = "https://api-inference.huggingface.co/models/Chanjeans/tfchatbot_2"
+HEADERS = {"Authorization": f"Bearer {HF_API_KEY}"}
 
-def chat_response(user_input, mode="emotion"):
+def chat_response(user_input, mode="emotion", max_retries=5):
     if mode not in ["emotion", "rational"]:
         raise HTTPException(status_code=400, detail="mode는 'emotion' 또는 'rational'이어야 합니다.")
 
     prompt = f"<{mode}><usr>{user_input}</usr><sys>"
+    payload = {
+        "inputs": prompt,
+        "parameters": {
+            "max_new_tokens": 128,
+            "temperature": 0.7,
+            "top_p": 0.9,
+            "top_k": 50,
+            "repetition_penalty": 1.2,
+            "do_sample": True
+        },
+        "options": {"wait_for_model": True}
+    }
 
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-
-    with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=128,
-            temperature=0.7,
-            top_p=0.9,
-            top_k=50,
-            repetition_penalty=1.2,
-            do_sample=True
-        )
-
-    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    # prompt 부분 제거 (불필요한 프롬프트까지 반환되지 않도록)
-    response_text = generated_text.replace(prompt, "").strip()
-
-    return response_text
-
+    for attempt in range(max_retries):
+        response = requests.post(API_URL, headers=HEADERS, json=payload)
+        if response.status_code == 200:
+            try:
+                result = response.json()
+                if isinstance(result, list) and "generated_text" in result[0]:
+                    generated_text = result[0]["generated_text"]
+                    return generated_text.replace(prompt, "").strip()
+                else:
+                    return "응답 형식이 예상과 다릅니다."
+            except Exception as e:
+                return f"JSON 파싱 오류: {e}"
+
+        elif response.status_code == 503:
+            # 모델 로딩 중
+            error_info = response.json()
+            estimated_time = error_info.get("estimated_time", 15)
+            time.sleep(min(estimated_time, 15))
+        else:
+            return f"API Error: {response.status_code}, {response.text}"
+
+    return "🚨 모델 로딩이 너무 오래 걸립니다. 잠시 후 다시 시도하세요."
 
 
 #우울분류 모델 추가
@@ -1357,13 +1371,4 @@ def chat_or_recommend(req: ChatOrRecommendRequest):
         if recommendation_msg:
             response_dict["recommendations"] = recommendations_list
 
-        return response_dict
-
-
-#def run_fastapi():
-#    uvicorn.run(app, host="0.0.0.0", port=7860)
-
-
-#if __name__ == "__main__":
-#    Thread(target=run_fastapi).start()
-#    iface.launch(server_name="0.0.0.0", server_port=7861)
+        return response_dict