Update app.py

app.py

@@ -74,9 +74,9 @@ class S3ModelLoader:
         s3_uri = self._get_s3_uri(model_name)
         try:
             logging.info(f"Trying to load {model_name} from S3...")
-            config = AutoConfig.from_pretrained(s3_uri, local_files_only=False)
-            model = AutoModelForCausalLM.from_pretrained(s3_uri, config=config, local_files_only=False)
-            tokenizer = AutoTokenizer.from_pretrained(s3_uri, config=config, local_files_only=False)
+            config = AutoConfig.from_pretrained(s3_uri, local_files_only=True)
+            model = AutoModelForCausalLM.from_pretrained(s3_uri, config=config, local_files_only=True)
+            tokenizer = AutoTokenizer.from_pretrained(s3_uri, config=config, local_files_only=True)
 
             if tokenizer.eos_token_id is not None and tokenizer.pad_token_id is None:
                 tokenizer.pad_token_id = config.pad_token_id or tokenizer.eos_token_id
@@ -87,10 +87,10 @@ class S3ModelLoader:
             logging.info(f"Model {model_name} not found in S3. Downloading...")
             try:
                 model_info = self.api.model_info(model_name)
-                files_to_download = [f.rfilename for f in self.api.list_repo_files(model_name)]
+                files_to_download = [f.filename for f in self.api.list_repo_files(model_name)]
                 
                 temp_dir = "temp_model"
-                os.makedirs(temp_dir, exist_ok = True)
+                os.makedirs(temp_dir, exist_ok=True)
                 
                 for file_name in files_to_download:
                     hf_hub_download(repo_id=model_name, filename=file_name, local_dir=temp_dir, token=HUGGINGFACE_HUB_TOKEN)