Tests: move generate tests to the right mixin and delete redundant tests (#34464)

* tmp commit

* tmp commit

* cull overwrites of deleted tests

* typo

* more specific docstring

* make fixup

* parameterize at the top?

* correction

* more deletions :D

* tmp commit

* for VLMs too

* fix _check_outputs

* test nit

* make fixup

* fix another flaky

* test_generate_from_inputs_embeds -- handle missing attention mask

tests/models/gemma/test_modeling_gemma.py

@@ -319,9 +319,6 @@ class GemmaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
     # This is because we are hitting edge cases with the causal_mask buffer
     model_split_percents = [0.5, 0.6]
 
-    # used in `test_torch_compile`
-    _torch_compile_test_ckpt = "google/gemma-2b"
-
     # used in `test_torch_compile_for_training`
     _torch_compile_train_cls = GemmaForCausalLM if is_torch_available() else None
 
@@ -419,51 +416,6 @@ class GemmaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
     def test_past_key_values_format(self):
         pass
 
-    @require_flash_attn
-    @require_torch_gpu
-    @pytest.mark.flash_attn_test
-    @slow
-    def test_flash_attn_2_generate_use_cache(self):
-        import torch
-
-        max_new_tokens = 30
-
-        for model_class in self.all_generative_model_classes:
-            config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
-
-            dummy_input = inputs_dict[model_class.main_input_name]
-            if dummy_input.dtype in [torch.float32, torch.bfloat16]:
-                dummy_input = dummy_input.to(torch.float16)
-
-            # make sure that all models have enough positions for generation
-            if hasattr(config, "max_position_embeddings"):
-                config.max_position_embeddings = max_new_tokens + dummy_input.shape[1] + 1
-
-            model = model_class(config)
-
-            with tempfile.TemporaryDirectory() as tmpdirname:
-                model.save_pretrained(tmpdirname)
-
-                dummy_attention_mask = inputs_dict.get("attention_mask", torch.ones_like(dummy_input))
-                # NOTE: Gemma apparently does not support right padding + use_cache with FA2.
-                dummy_attention_mask[:, -1] = 1
-
-                model = model_class.from_pretrained(
-                    tmpdirname,
-                    torch_dtype=torch.float16,
-                    attn_implementation="flash_attention_2",
-                    low_cpu_mem_usage=True,
-                ).to(torch_device)
-
-                # Just test that a large cache works as expected
-                _ = model.generate(
-                    dummy_input,
-                    attention_mask=dummy_attention_mask,
-                    max_new_tokens=max_new_tokens,
-                    do_sample=False,
-                    use_cache=True,
-                )
-
     @require_flash_attn
     @require_torch_gpu
     @pytest.mark.flash_attn_test