Fixed patch

main_backend_lighteval.py

@@ -3,7 +3,6 @@ import pprint
 
 import lighteval.models.endpoint_model
 class GoodInferenceEndpointModel(lighteval.models.endpoint_model.InferenceEndpointModel):
-
     @property 
     def add_special_tokens(self):
         return True
@@ -11,10 +10,11 @@ class GoodInferenceEndpointModel(lighteval.models.endpoint_model.InferenceEndpoi
     def greedy_until(self, requests: list, *args, **kwargs):
         for request in requests:
             request.tokenized_context = self.tok_encode(request.context)
-        super().greedy_until(requests, *args, **kwargs)
+        # using this and not super() because we don't want self to change
+        return lighteval.models.endpoint_model.InferenceEndpointModel.greedy_until(self, requests, *args, **kwargs)
 
     def __process_batch_generate(self, requests: list, returns_logits: bool):
-        return super().__process_batch_generate(requests)
+        return lighteval.models.endpoint_model.InferenceEndpointModel.__process_batch_generate(self, requests)
 
     @property
     def disable_tqdm(self) -> bool: