diff --git a/torchtitan/parallelisms/parallelize_llama.py b/torchtitan/parallelisms/parallelize_llama.py
index 5440bbe58..4d4c60bc0 100644
--- a/torchtitan/parallelisms/parallelize_llama.py
+++ b/torchtitan/parallelisms/parallelize_llama.py
@@ -238,8 +238,8 @@ def apply_tp(
     torch.ops._c10d_functional.reduce_scatter_tensor.default,
     # for low precision training, it's useful to always save
     # the result of max(abs(tensor))
-    # torch.ops.aten.abs.default,
-    # torch.ops.aten.max.default,
+    torch.ops.aten.abs.default,
+    torch.ops.aten.max.default,
 }