diff --git a/plugins/attention-and-distributed-packing/src/fms_acceleration_aadp/framework_plugin_padding_free.py b/plugins/attention-and-distributed-packing/src/fms_acceleration_aadp/framework_plugin_padding_free.py
index 65ba520b..1141c57d 100644
--- a/plugins/attention-and-distributed-packing/src/fms_acceleration_aadp/framework_plugin_padding_free.py
+++ b/plugins/attention-and-distributed-packing/src/fms_acceleration_aadp/framework_plugin_padding_free.py
@@ -21,11 +21,23 @@
 from peft import LoraConfig
 from transformers import DataCollatorForSeq2Seq, TrainingArguments
 from transformers.trainer_utils import RemoveColumnsCollator
-from trl import (  # pylint: disable=import-error, no-name-in-module
-    DataCollatorForCompletionOnlyLM,
-)
 import torch
 
+# Handle trl version compatibility
+# In trl < 0.19: DataCollatorForCompletionOnlyLM
+# In trl >= 0.19: May be renamed or moved
+try:
+    # pylint: disable=import-error, no-name-in-module
+    from trl import DataCollatorForCompletionOnlyLM
+except ImportError:
+    # Fallback for newer trl versions where it might be renamed
+    try:
+        from trl.trainer.utils import DataCollatorForCompletionOnlyLM
+    except ImportError:
+        # If still not available, create a placeholder that will never match
+        # This allows the plugin to load even if this specific collator isn't used
+        DataCollatorForCompletionOnlyLM = type('DataCollatorForCompletionOnlyLM', (), {})
+
 
 class PaddingFreeAccelerationPlugin(AccelerationPlugin):