diff --git a/training/configs/external_generalisation/math_curriculum_qwen_7b.yaml b/training/configs/external_generalisation/math_curriculum_qwen_7b.yaml
index 0a32a24c..1cc5ee57 100644
--- a/training/configs/external_generalisation/math_curriculum_qwen_7b.yaml
+++ b/training/configs/external_generalisation/math_curriculum_qwen_7b.yaml
@@ -94,6 +94,8 @@ actor_rollout_ref:
     ppo_epochs: 1
     shuffle: False
     ulysses_sequence_parallel_size: 1 # sp size
+    checkpoint:
+      contents: ['model', 'optimizer', 'extra']
     optim:
       lr: 1e-6
       lr_warmup_steps_ratio: 0.  # the total steps will be injected during runtime