Mark 2 tests as flaky for now (#37038)

* fix

* fix

* fix

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

tests/models/smolvlm/test_modeling_smolvlm.py

@@ -29,6 +29,7 @@ from transformers import (
 )
 from transformers.testing_utils import (
     cleanup,
+    is_flaky,
     require_torch,
     require_torch_sdpa,
     slow,
@@ -373,6 +374,11 @@ class SmolVLMForConditionalGenerationModelTest(GenerationTesterMixin, ModelTeste
     def test_prompt_lookup_decoding_matches_greedy_search(self):
         pass
 
+    @pytest.mark.generate
+    @is_flaky(description="TODO: check why flaky")
+    def test_generate_methods_with_logits_to_keep(self):
+        super().test_generate_methods_with_logits_to_keep()
+
     @unittest.skip(reason=" FlashAttention only support fp16 and bf16 data type")
     def test_flash_attn_2_fp32_ln(self):
         pass

tests/models/zamba/test_modeling_zamba.py

@@ -22,6 +22,7 @@ import pytest
 
 from transformers import AutoTokenizer, ZambaConfig, is_torch_available
 from transformers.testing_utils import (
+    is_flaky,
     require_bitsandbytes,
     require_flash_attn,
     require_torch,
@@ -327,6 +328,7 @@ class ZambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
         config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
         self.model_tester.create_and_check_decoder_model_past_large_inputs(*config_and_inputs)
 
+    @is_flaky(description="TODO: ydshieh")
     def test_initialization(self):
         r"""
         Overriding the test_initialization test as the A_log and D params of the Mamba block are initialized differently