[generate] remove cache v4.47 deprecations (#36212)

tests/models/phimoe/test_modeling_phimoe.py

@@ -22,6 +22,7 @@ from parameterized import parameterized
 
 from transformers import PhimoeConfig, StaticCache, is_torch_available, set_seed
 from transformers.testing_utils import (
+    is_flaky,
     require_torch,
     slow,
     torch_device,
@@ -449,6 +450,7 @@ class PhimoeModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix
         self.assertFalse(torch.allclose(original_long_output, scaled_long_output, atol=1e-5))
 
     @parameterized.expand([("longrope",)])
+    @is_flaky()  # TODO (joao): unify rope tests in the mixin
     def test_model_rope_scaling_short_long_factor(self, scaling_type):
         config, _ = self.model_tester.prepare_config_and_inputs_for_common()
         n_factors = config.hidden_size // config.num_key_value_heads // 2

tests/models/qwen2_5_vl/test_modeling_qwen2_5_vl.py

@@ -27,6 +27,7 @@ from transformers import (
     is_vision_available,
 )
 from transformers.testing_utils import (
+    is_flaky,
     require_flash_attn,
     require_torch,
     require_torch_gpu,
@@ -347,6 +348,10 @@ class Qwen2_5_VLModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.Test
     def test_generate_compile_fullgraph(self):
         pass
 
+    @is_flaky()  # TODO (joao/raushan): Investigate why this test is flaky on this model
+    def test_prompt_lookup_decoding_matches_greedy_search(self):
+        super().test_prompt_lookup_decoding_matches_greedy_search()
+
 
 @require_torch
 class Qwen2_5_VLIntegrationTest(unittest.TestCase):