Upload app.py

app.py

@@ -31,7 +31,8 @@ class CustomInferenceClient(LLM, KwArgsModel):
     inference_client: InferenceClient
 
     def __init__(self, model_name: str, hf_token: str, kwargs: Optional[Dict[str, Any]] = None):
-        inference_client = InferenceClient(model=model_name, token=hf_token)
+        # inference_client = InferenceClient(model=model_name, token=hf_token)
+        inference_client = InferenceClient(model=model_name, token=os.getenv("HUGGINGFACEHUB_API_TOKEN"))
         super().__init__(
             model_name=model_name,
             hf_token=hf_token,