fixed recursion

main_backend_lighteval.py

@@ -2,7 +2,10 @@ import logging
 import pprint
 
 import lighteval.models.endpoint_model
+orig_endpoint_model_greedy_until = lighteval.models.endpoint_model.InferenceEndpointModel.greedy_until
+orig_endpoint_model_process_batch_generate = lighteval.models.endpoint_model.InferenceEndpointModel.__process_batch_generate
 class GoodInferenceEndpointModel(lighteval.models.endpoint_model.InferenceEndpointModel):
+
     @property 
     def add_special_tokens(self):
         return True
@@ -11,10 +14,10 @@ class GoodInferenceEndpointModel(lighteval.models.endpoint_model.InferenceEndpoi
         for request in requests:
             request.tokenized_context = self.tok_encode(request.context)
         # 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)
+        return orig_endpoint_model_greedy_until(self, requests, *args, **kwargs)
 
     def __process_batch_generate(self, requests: list, returns_logits: bool):
-        return lighteval.models.endpoint_model.InferenceEndpointModel.__process_batch_generate(self, requests)
+        return orig_endpoint_model_process_batch_generate(self, requests)
 
     @property
     def disable_tqdm(self) -> bool: