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/phi/test_modeling_phi.py

@@ -17,15 +17,11 @@
 
 import unittest
 
-import pytest
 from parameterized import parameterized
 
 from transformers import PhiConfig, is_torch_available, set_seed
 from transformers.testing_utils import (
-    require_bitsandbytes,
-    require_flash_attn,
     require_torch,
-    require_torch_gpu,
     slow,
     torch_device,
 )
@@ -468,43 +464,6 @@ class PhiModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin,
             torch.testing.assert_close(ntk_sin_long, original_sin_long)
         self.assertTrue((ntk_scaling_rope.inv_freq <= original_rope.inv_freq).all())
 
-    @require_flash_attn
-    @require_torch_gpu
-    @require_bitsandbytes
-    @pytest.mark.flash_attn_test
-    @slow
-    # Copied from tests.models.llama.test_modeling_llama.LlamaModelTest.test_flash_attn_2_generate_padding_right with LlamaForCausalLM->PhiForCausalLM,LlamaTokenizer->AutoTokenizer,meta-llama/Llama-2-7b-hf->microsoft/phi-1
-    def test_flash_attn_2_generate_padding_right(self):
-        """
-        Overwritting the common test as the test is flaky on tiny models
-        """
-        model = PhiForCausalLM.from_pretrained(
-            "microsoft/phi-1",
-            load_in_4bit=True,
-            device_map={"": 0},
-        )
-
-        tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-1")
-
-        texts = ["hi", "Hello this is a very long sentence"]
-
-        tokenizer.padding_side = "right"
-        tokenizer.pad_token = tokenizer.eos_token
-
-        inputs = tokenizer(texts, return_tensors="pt", padding=True).to(0)
-
-        output_native = model.generate(**inputs, max_new_tokens=20, do_sample=False)
-        output_native = tokenizer.batch_decode(output_native)
-
-        model = PhiForCausalLM.from_pretrained(
-            "microsoft/phi-1", load_in_4bit=True, device_map={"": 0}, attn_implementation="flash_attention_2"
-        )
-
-        output_fa_2 = model.generate(**inputs, max_new_tokens=20, do_sample=False)
-        output_fa_2 = tokenizer.batch_decode(output_fa_2)
-
-        self.assertListEqual(output_native, output_fa_2)
-
 
 @slow
 @require_torch