Make tiny model creation + pipeline testing more robust (#22500)

* Final Tiny things

---------

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

tests/models/led/test_modeling_led.py

@@ -294,6 +294,15 @@ class LEDModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin,
     test_missing_keys = False
     test_torchscript = False
 
+    # TODO: Fix the failed tests when this model gets more usage
+    def is_pipeline_test_to_skip(
+        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+    ):
+        if pipeline_test_casse_name == "QAPipelineTests" and not tokenizer_name.endswith("Fast"):
+            return True
+
+        return False
+
     def setUp(self):
         self.model_tester = LEDModelTester(self)
         self.config_tester = ConfigTester(self, config_class=LEDConfig)

tests/models/nllb_moe/test_modeling_nllb_moe.py

@@ -265,6 +265,13 @@ class NllbMoeModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
     test_missing_keys = True
     test_torchscript = False
 
+    # TODO: Fix the failed tests when this model gets more usage
+    def is_pipeline_test_to_skip(
+        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+    ):
+        # Saving the slow tokenizer after saving the fast tokenizer causes the loading of the later hanging forever.
+        return True
+
     def setUp(self):
         self.model_tester = NllbMoeModelTester(self)
         self.config_tester = ConfigTester(self, config_class=NllbMoeConfig)

tests/models/splinter/test_modeling_splinter.py

@@ -230,6 +230,8 @@ class SplinterModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
     ):
         if pipeline_test_casse_name == "QAPipelineTests":
             return True
+        elif pipeline_test_casse_name == "FeatureExtractionPipelineTests" and tokenizer_name.endswith("Fast"):
+            return True
 
         return False