Update app.py

app.py

@@ -3,36 +3,42 @@ from dotenv import load_dotenv
 import gradio as gr
 from huggingface_hub import InferenceClient
 import pandas as pd
-from typing import List, Tuple
 import json
 from datetime import datetime
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
-import spaces
+import tempfile
 
 # 환경 변수 설정
 HF_TOKEN = os.getenv("HF_TOKEN")
 MODEL_ID = "CohereForAI/c4ai-command-r-plus-08-2024"
 
-os.environ["TRANSFORMERS_CACHE"] = "/persistent/transformers_cache"
-os.environ["TORCH_HOME"] = "/persistent/torch_cache"
-os.environ["HF_HOME"] = "/persistent/huggingface"
+# 임시 디렉토리 설정
+TEMP_DIR = tempfile.mkdtemp()
+os.environ["TRANSFORMERS_CACHE"] = os.path.join(TEMP_DIR, "transformers_cache")
+os.environ["TORCH_HOME"] = os.path.join(TEMP_DIR, "torch_cache")
+os.environ["HF_HOME"] = os.path.join(TEMP_DIR, "huggingface")
 
 class ModelManager:
     def __init__(self):
-        self.cache_dir = "/persistent/model_cache"
+        self.cache_dir = os.path.join(TEMP_DIR, "model_cache")
+        os.makedirs(self.cache_dir, exist_ok=True)
         self.model = None
         self.tokenizer = None
         self.setup_model()
     
     def setup_model(self):
         try:
+            print("토크나이저 로딩 시작...")
             self.tokenizer = AutoTokenizer.from_pretrained(
                 MODEL_ID,
                 token=HF_TOKEN,
                 trust_remote_code=True,
                 cache_dir=self.cache_dir
             )
+            print("토크나이저 로딩 완료")
+            
+            print("모델 로딩 시작...")
             self.model = AutoModelForCausalLM.from_pretrained(
                 MODEL_ID,
                 token=HF_TOKEN,
@@ -42,8 +48,21 @@ class ModelManager:
                 low_cpu_mem_usage=True,
                 cache_dir=self.cache_dir
             )
+            print("모델 로딩 완료")
         except Exception as e:
-            raise Exception(f"Model loading failed: {e}")
+            print(f"모델 로딩 중 오류 발생: {e}")
+            try:
+                print("기본 설정으로 다시 시도...")
+                self.model = AutoModelForCausalLM.from_pretrained(
+                    MODEL_ID,
+                    token=HF_TOKEN,
+                    torch_dtype=torch.float16,
+                    trust_remote_code=True,
+                    cache_dir=self.cache_dir
+                )
+                print("기본 설정으로 모델 로딩 완료")
+            except Exception as e2:
+                raise Exception(f"모델 로딩 완전히 실패: {e2}")
 
 class ChatHistory:
     def __init__(self):