diff --git a/train_configs/llama2_13b.toml b/train_configs/llama2_13b.toml
index 280ac2ae..f3048ac4 100644
--- a/train_configs/llama2_13b.toml
+++ b/train_configs/llama2_13b.toml
@@ -33,11 +33,13 @@ max_norm = 1.0  # grad norm clipping
 steps = 1000
 data_parallel_degree = -1
 tensor_parallel_degree = 1
-pipeline_parallel_degree = 1
 fp8_linear = ""
 compile = false
 dataset = "c4"
 
+[experimental]
+pipeline_parallel_degree = 1
+
 [checkpoint]
 enable_checkpoint = false
 folder = "checkpoint"
diff --git a/train_configs/llama2_70b.toml b/train_configs/llama2_70b.toml
index 959c270a..97b1bc71 100644
--- a/train_configs/llama2_70b.toml
+++ b/train_configs/llama2_70b.toml
@@ -33,11 +33,13 @@ max_norm = 1.0  # grad norm clipping
 steps = 1000
 data_parallel_degree = -1
 tensor_parallel_degree = 8  # 8-way TP
-pipeline_parallel_degree = 1
 fp8_linear = ""
 compile = false
 dataset = "c4"
 
+[experimental]
+pipeline_parallel_degree = 1
+
 [checkpoint]
 enable_checkpoint = false
 folder = "checkpoint"
diff --git a/train_configs/llama2_7b.toml b/train_configs/llama2_7b.toml
index f2e66de7..95b4c496 100644
--- a/train_configs/llama2_7b.toml
+++ b/train_configs/llama2_7b.toml
@@ -32,11 +32,13 @@ max_norm = 1.0  # grad norm clipping
 steps = 1000
 data_parallel_degree = -1
 tensor_parallel_degree = 1  # dp-only would be sufficient for 7B
-pipeline_parallel_degree = 1
 fp8_linear = ""
 compile = false
 dataset = "c4"
 
+[experimental]
+pipeline_parallel_degree = 1
+
 [checkpoint]
 enable_checkpoint = false
 folder = "checkpoint"
diff --git a/train_configs/llama3_70b.toml b/train_configs/llama3_70b.toml
index f45632ad..d498e677 100644
--- a/train_configs/llama3_70b.toml
+++ b/train_configs/llama3_70b.toml
@@ -33,11 +33,13 @@ max_norm = 1.0  # grad norm clipping
 steps = 1000
 data_parallel_degree = -1
 tensor_parallel_degree = 8  # 8-way TP
-pipeline_parallel_degree = 1
 fp8_linear = ""
 compile = false
 dataset = "c4"
 
+[experimental]
+pipeline_parallel_degree = 1
+
 [checkpoint]
 enable_checkpoint = false
 folder = "checkpoint"
diff --git a/train_configs/llama3_8b.toml b/train_configs/llama3_8b.toml
index aaba99a2..f194addb 100644
--- a/train_configs/llama3_8b.toml
+++ b/train_configs/llama3_8b.toml
@@ -33,11 +33,13 @@ max_norm = 1.0  # grad norm clipping
 steps = 1000
 data_parallel_degree = -1
 tensor_parallel_degree = 1
-pipeline_parallel_degree = 1
 fp8_linear = ""
 compile = false
 dataset = "c4"
 
+[experimental]
+pipeline_parallel_degree = 1
+
 [checkpoint]
 enable_checkpoint = false
 folder = "checkpoint"