Fix CI issues (#35662)

* make explicit gpu dep

* [run-slow] bamba

tests/models/bamba/test_modeling_bamba.py

@@ -22,6 +22,7 @@ import pytest
 from transformers import AutoTokenizer, BambaConfig, is_torch_available
 from transformers.testing_utils import (
     require_torch,
+    require_torch_gpu,
     slow,
     torch_device,
 )
@@ -487,6 +488,7 @@ class BambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
 
 @slow
 @require_torch
+@require_torch_gpu
 class BambaModelIntegrationTest(unittest.TestCase):
     model = None
     tokenizer = None