Remove head mask in generative models (#35786)

* just squash into one commit

* delete print

tests/generation/test_utils.py

@@ -1414,6 +1414,7 @@ class GenerationTesterMixin:
         attention_names = ["encoder_attentions", "decoder_attentions", "cross_attentions"]
         for model_class in self.all_generative_model_classes:
             config, inputs_dict = self.prepare_config_and_inputs_for_generate()
+            config._attn_implementation = "eager"  # head mask works only in eager mode and will be removed soon
             text_config = config.get_text_config()
             if self.has_attentions:
                 config._attn_implementation = "eager"  # can't output attentions otherwise