try secrets

handler.py

@@ -6,8 +6,9 @@ import torch
 
 class EndpointHandler():
     def __init__(self, path=""):
+        auth_token = os.environ.get("LLAMA_TOK")
         device = "cuda" if torch.cuda.is_available() else "cpu"
-        model = NatureLM.from_pretrained("EarthSpeciesProject/NatureLM-audio", device=device)
+        model = NatureLM.from_pretrained("EarthSpeciesProject/NatureLM-audio", device=device, hf_auth_token=auth_token)
         self.model = model.eval().to(device)
         self.pipeline = Pipeline(model=self.model, cfg_path="inference.yml")
 
@@ -51,7 +52,7 @@ class EndpointHandler():
                 input_sample_rate=sample_rate
             )
             
-            return [{"result": results[0], "query": query}]
+            return [{"result": results[0], "query": query, "token": os.environ.get("LLAMA_TOK")}]
             
         except Exception as e:
             return [{"error": f"Error processing audio: {str(e)}"}]