From 727e0e51a0e73e2c78b94314e7d2099e9bad2a5a Mon Sep 17 00:00:00 2001 From: Lucas Jia Date: Thu, 21 May 2026 12:27:14 -0700 Subject: [PATCH 1/2] fix: remove preset_reward_function from RLVR hyperparameters to prevent empty value being sent to service --- sagemaker-train/src/sagemaker/train/rlvr_trainer.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/sagemaker-train/src/sagemaker/train/rlvr_trainer.py b/sagemaker-train/src/sagemaker/train/rlvr_trainer.py index 49b35f124e..333a93fc55 100644 --- a/sagemaker-train/src/sagemaker/train/rlvr_trainer.py +++ b/sagemaker-train/src/sagemaker/train/rlvr_trainer.py @@ -171,6 +171,9 @@ def _process_hyperparameters(self): if hasattr(self.hyperparameters, 'reward_lambda_arn'): delattr(self.hyperparameters, 'reward_lambda_arn') self.hyperparameters._specs.pop('reward_lambda_arn', None) + if hasattr(self.hyperparameters, 'preset_reward_function'): + delattr(self.hyperparameters, 'preset_reward_function') + self.hyperparameters._specs.pop('preset_reward_function', None) if hasattr(self.hyperparameters, 'data_path'): delattr(self.hyperparameters, 'data_path') self.hyperparameters._specs.pop('data_path', None) From e8148dc9d8a554a8275220de07a14bded22733e1 Mon Sep 17 00:00:00 2001 From: Lucas Jia Date: Thu, 21 May 2026 13:45:47 -0700 Subject: [PATCH 2/2] unskip test_rlvr_trainer_lora_complete_workflow for verification --- .../tests/integ/train/test_rlvr_trainer_integration.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py b/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py index 63d3ae3134..3cf6851a98 100644 --- a/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py +++ b/sagemaker-train/tests/integ/train/test_rlvr_trainer_integration.py @@ -22,7 +22,7 @@ from sagemaker.train.common import TrainingType -@pytest.mark.skip(reason="Skipping GPU resource intensive test") +# @pytest.mark.skip(reason="Skipping GPU resource intensive test") def test_rlvr_trainer_lora_complete_workflow(sagemaker_session): """Test complete RLVR training workflow with LORA."""