CI hotfix: xfail test_training_with_transformers_paged (#4046)

Author: albertvillanova

tests/slow/test_grpo_slow.py

@@ -19,8 +19,10 @@
 import numpy as np
 import pytest
 import torch
+import transformers
 from accelerate.utils.memory import release_memory
 from datasets import Dataset, Features, Image, Value, load_dataset
+from packaging.version import Version
 from parameterized import parameterized
 from transformers import (
     AutoModelForCausalLM,
@@ -171,6 +173,8 @@ def test_training_with_liger_grpo_loss_and_peft(self, model_name):
     @parameterized.expand(MODELS_TO_TEST)
     def test_training_with_transformers_paged(self, model_name):
         """Test that training works with transformers paged implementation (requires GPU)."""
+        if Version(transformers.__version__) < Version("4.56.2"):
+            pytest.xfail("Upstream bug in transformers (GH#40692). Fix merged; awaiting release >= 4.56.2")
         training_args = GRPOConfig(
             output_dir=self.tmp_dir,
             learning_rate=0.1,  # increase the learning rate to speed up the test

tests/test_online_dpo_trainer.py

@@ -12,9 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-
 import pytest
+import transformers
 from datasets import Dataset, features, load_dataset
+from packaging.version import Version
 from parameterized import parameterized
 from transformers import AutoModelForCausalLM, AutoModelForSequenceClassification, AutoTokenizer
 from transformers.testing_utils import require_peft, require_torch_accelerator, require_vision
@@ -421,6 +422,8 @@ def test_generation_config_setup(self):
     @require_torch_accelerator
     @parameterized.expand([("standard_prompt_only",), ("conversational_prompt_only",)])
     def test_training_with_transformers_paged(self, config_name):
+        if Version(transformers.__version__) < Version("4.56.2"):
+            pytest.xfail("Upstream bug in transformers (GH#40692). Fix merged; awaiting release >= 4.56.2")
         training_args = OnlineDPOConfig(
             output_dir=self.tmp_dir,
             per_device_train_batch_size=2,