update

eagle_vl/serve/inference.py

@@ -20,13 +20,14 @@ logger = logging.getLogger(__name__)
 
 def load_model(model_path: str = "NVEagle/Eagle2.5-VL-8B-Preview"):
 
+    token = os.environ.get("HF_TOKEN")
     # hotfix the model to use flash attention 2
-    config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
+    config = AutoConfig.from_pretrained(model_path, trust_remote_code=True, use_auth_token=token)
     config._attn_implementation = "flash_attention_2"
     config.vision_config._attn_implementation = "flash_attention_2"
     config.text_config._attn_implementation = "flash_attention_2"
     print("Successfully set the attn_implementation to flash_attention_2")
-    token = os.environ.get("HF_TOKEN")
+
     logger.info(f"token = {token[:4]}***{token[-2:]}")
     model = AutoModel.from_pretrained(
         model_path,