Generate: add test to check KV format (#23403)

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

tests/models/bloom/test_modeling_bloom.py

@@ -393,6 +393,10 @@ class BloomModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_bloom_weight_initialization(*config_and_inputs)
 
+    @unittest.skip("Bloom has a non-standard KV cache format.")
+    def test_past_key_values_format(self):
+        pass
+
     @slow
     def test_model_from_pretrained(self):
         for model_name in BLOOM_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:

tests/models/gpt_bigcode/test_modeling_gpt_bigcode.py

@@ -475,6 +475,10 @@ class GPTBigCodeModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTeste
     def test_disk_offload(self):
         pass
 
+    @unittest.skip("BigCodeGPT has a non-standard KV cache format.")
+    def test_past_key_values_format(self):
+        pass
+
     def test_gpt_bigcode_model(self):
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_gpt_bigcode_model(*config_and_inputs)

tests/models/reformer/test_modeling_reformer.py

@@ -831,8 +831,12 @@ class ReformerLSHAttnModelTest(
                 [expected_shape] * len(iter_hidden_states),
             )
 
+    @unittest.skip("Fails because the sequence length is not a multiple of 4")
     def test_problem_types(self):
-        # Fails because the sequence length is not a multiple of 4
+        pass
+
+    @unittest.skip("Fails because the sequence length is not a multiple of 4")
+    def test_past_key_values_format(self):
         pass