diff --git a/egs/aishell/ASR/whisper/train.py b/egs/aishell/ASR/whisper/train.py
index 073b23713..6ccb8d363 100755
--- a/egs/aishell/ASR/whisper/train.py
+++ b/egs/aishell/ASR/whisper/train.py
@@ -793,7 +793,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)
 
diff --git a/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/do_not_use_it_directly.py b/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/do_not_use_it_directly.py
index 8e16cc0bf..a3f387636 100755
--- a/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/do_not_use_it_directly.py
+++ b/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/do_not_use_it_directly.py
@@ -1035,7 +1035,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)
 
diff --git a/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/finetune.py b/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/finetune.py
index 47c4ed312..81c69e5e0 100755
--- a/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/finetune.py
+++ b/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/finetune.py
@@ -1119,7 +1119,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)
 
diff --git a/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/train.py b/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/train.py
index dc5d0a858..728104580 100755
--- a/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/train.py
+++ b/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/train.py
@@ -1035,7 +1035,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)
 
diff --git a/egs/multi_zh-hans/ASR/whisper/train.py b/egs/multi_zh-hans/ASR/whisper/train.py
index 11a22eec1..b1b60077c 100644
--- a/egs/multi_zh-hans/ASR/whisper/train.py
+++ b/egs/multi_zh-hans/ASR/whisper/train.py
@@ -824,7 +824,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)
 
diff --git a/egs/wenetspeech/ASR/whisper/train.py b/egs/wenetspeech/ASR/whisper/train.py
index 4b7c1ca42..6ff500ab9 100644
--- a/egs/wenetspeech/ASR/whisper/train.py
+++ b/egs/wenetspeech/ASR/whisper/train.py
@@ -803,7 +803,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)
 
diff --git a/egs/wenetspeech/KWS/zipformer/train.py b/egs/wenetspeech/KWS/zipformer/train.py
index b5cf3359a..eddec7303 100755
--- a/egs/wenetspeech/KWS/zipformer/train.py
+++ b/egs/wenetspeech/KWS/zipformer/train.py
@@ -1187,7 +1187,7 @@ def run(rank, world_size, args):
 
     if params.print_diagnostics:
         opts = diagnostics.TensorDiagnosticOptions(
-            2**22
+            512
         )  # allow 4 megabytes per sub-module
         diagnostic = diagnostics.attach_diagnostics(model, opts)