Update app.py

app.py

@@ -11,6 +11,10 @@ import spaces
 import time
 import subprocess
 
+from transformers_modules.apple.OpenELM-3B-Instruct.d3c76da586450c73898e6bed70cecf3376300fb3.configuration_openelm import OpenELMConfig
+
+
+
 # flash-attn 라이브러리 설치. CUDA 빌드는 건너뜀.
 subprocess.run(
     "pip install flash-attn --no-build-isolation",
@@ -29,7 +33,14 @@ model = AutoModelForCausalLM.from_pretrained(
     token=token,
     trust_remote_code=True,
 )
-tok = AutoTokenizer.from_pretrained("apple/OpenELM-3B-Instruct", token=token, trust_remote_code=True)
+# bug fix
+# tok = AutoTokenizer.from_pretrained("apple/OpenELM-3B-Instruct", token=token, trust_remote_code=True)
+tok = AutoTokenizer.from_pretrained(
+    "apple/OpenELM-3B-Instruct",
+    token=token,
+    trust_remote_code=True,
+    config=OpenELMConfig(),
+)
 
 # 종료 토큰 ID 설정
 terminators = [