diff --git a/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary-8k.yaml b/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary-8k.yaml
index 0a5a588..09fdee9 100644
--- a/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary-8k.yaml
+++ b/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary-8k.yaml
@@ -11,7 +11,7 @@ model:
     mlp_checkpoint_lvl: 0
 
 datamodule:
-  batch_size: ${eval:"1 if ${train.gpu_mem} < 24 else (2 if ${train.gpu_mem} < 40 else (4 if ${train.gpu} < 80 else 8))"}
+  batch_size: ${eval:"1 if ${train.gpu_mem} < 24 else (2 if ${train.gpu_mem} < 40 else (4 if ${train.gpu_mem} < 80 else 8))"}
 
 train:
   optimizer:
diff --git a/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary.yaml b/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary.yaml
index 79c3730..d5caafd 100644
--- a/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary.yaml
+++ b/training/configs/experiment/pile/gpt3-2.7B-flash-hdim128-rotary.yaml
@@ -11,7 +11,7 @@ model:
     mlp_checkpoint_lvl: 0
 
 datamodule:
-  batch_size: ${eval:"4 if ${train.gpu_mem} < 24 else (8 if ${train.gpu_mem} < 40 else (16 if ${train.gpu} < 80 else 32))"}
+  batch_size: ${eval:"4 if ${train.gpu_mem} < 24 else (8 if ${train.gpu_mem} < 40 else (16 if ${train.gpu_mem} < 80 else 32))"}
 
 train:
   optimizer:
diff --git a/training/configs/experiment/pile/gpt3-2.7B-flash-rotary-8k.yaml b/training/configs/experiment/pile/gpt3-2.7B-flash-rotary-8k.yaml
index 4584e6b..b259a29 100644
--- a/training/configs/experiment/pile/gpt3-2.7B-flash-rotary-8k.yaml
+++ b/training/configs/experiment/pile/gpt3-2.7B-flash-rotary-8k.yaml
@@ -11,7 +11,7 @@ model:
     mlp_checkpoint_lvl: 0
 
 datamodule:
-  batch_size: ${eval:"1 if ${train.gpu_mem} < 24 else (2 if ${train.gpu_mem} < 40 else (4 if ${train.gpu} < 80 else 8))"}
+  batch_size: ${eval:"1 if ${train.gpu_mem} < 24 else (2 if ${train.gpu_mem} < 40 else (4 if ${train.gpu_mem} < 80 else 8))"}
 
 train:
   optimizer:
diff --git a/training/configs/experiment/pile/gpt3-2.7B-flash-rotary.yaml b/training/configs/experiment/pile/gpt3-2.7B-flash-rotary.yaml
index c16ab33..1e1684c 100644
--- a/training/configs/experiment/pile/gpt3-2.7B-flash-rotary.yaml
+++ b/training/configs/experiment/pile/gpt3-2.7B-flash-rotary.yaml
@@ -11,7 +11,7 @@ model:
     mlp_checkpoint_lvl: 0
 
 datamodule:
-  batch_size: ${eval:"4 if ${train.gpu_mem} < 24 else (8 if ${train.gpu_mem} < 40 else (16 if ${train.gpu} < 80 else 32))"}
+  batch_size: ${eval:"4 if ${train.gpu_mem} < 24 else (8 if ${train.gpu_mem} < 40 else (16 if ${train.gpu_mem} < 80 else 32))"}
 
 train:
   optimizer: