diff --git a/demo/finetune_classifier.py b/demo/finetune_classifier.py
index 4e9540d143e2d7bd407cf852a520e3c2c3a3135e..ce14d3ef5116cbf7b01ea2f1ec0a0257870f9a40 100644
--- a/demo/finetune_classifier.py
+++ b/demo/finetune_classifier.py
@@ -177,15 +177,15 @@ if args.use_lr_decay:
         lr_scheduler,
         parameters=model.parameters(),
         weight_decay=args.wd,
-        apply_decay_param_fun=lambda n: param_name_to_exclue_from_weight_decay.match(n),
+        apply_decay_param_fun=lambda n: not param_name_to_exclue_from_weight_decay.match(n),
         grad_clip=g_clip)
 else:
     lr_scheduler = None
-    opt = P.optimizer.Adam(
+    opt = P.optimizer.AdamW(
         args.lr,
         parameters=model.parameters(),
         weight_decay=args.wd,
-        apply_decay_param_fun=lambda n: param_name_to_exclue_from_weight_decay.match(n),
+        apply_decay_param_fun=lambda n: not param_name_to_exclue_from_weight_decay.match(n),
         grad_clip=g_clip)
 
 scaler = P.amp.GradScaler(enable=args.use_amp)
@@ -209,7 +209,8 @@ with LogWriter(
                 lr_scheduler and lr_scheduler.step()
 
                 if step % 10 == 0:
-                    _lr = lr_scheduler.get_lr()
+                    _lr = lr_scheduler.get_lr(
+                    ) if args.use_lr_decay else args.lr
                     if args.use_amp:
                         _l = (loss / scaler._scale).numpy()
                         msg = '[step-%d] train loss %.5f lr %.3e scaling %.3e' % (
diff --git a/demo/finetune_classifier_distributed.py b/demo/finetune_classifier_distributed.py
index d1df8675369125baf3c1605347cfa0848b0903de..d4b1195d0664a716ebf5b9bc03df251698f660f5 100644
--- a/demo/finetune_classifier_distributed.py
+++ b/demo/finetune_classifier_distributed.py
@@ -144,7 +144,7 @@ lr_scheduler = P.optimizer.lr.LambdaDecay(
 opt = P.optimizer.AdamW(
     learning_rate=lr_scheduler,
     parameters=model.parameters(),
-    apply_decay_param_fun=lambda n: param_name_to_exclue_from_weight_decay.match(n),
+    apply_decay_param_fun=lambda n: not param_name_to_exclue_from_weight_decay.match(n),
     weight_decay=args.wd,
     grad_clip=g_clip)
 scaler = P.amp.GradScaler(enable=args.use_amp)
diff --git a/demo/finetune_ner.py b/demo/finetune_ner.py
index 7489f16dac0b236bd703bd20565cf0500b5f48ab..6929afdcb2492f58cdaa7c3ecdb5140fe0b843e2 100644
--- a/demo/finetune_ner.py
+++ b/demo/finetune_ner.py
@@ -210,7 +210,7 @@ opt = P.optimizer.AdamW(
     lr_scheduler,
     parameters=model.parameters(),
     weight_decay=args.wd,
-    apply_decay_param_fun=lambda n: param_name_to_exclue_from_weight_decay.match(n),
+    apply_decay_param_fun=lambda n: not param_name_to_exclue_from_weight_decay.match(n),
     grad_clip=g_clip)
 
 scaler = P.amp.GradScaler(enable=args.use_amp)
diff --git a/demo/finetune_sentiment_analysis.py b/demo/finetune_sentiment_analysis.py
index 015d29d4d2fe0acdf34d0032ebdffd3946515e62..16087fa0a59bd6fbfa1cfd460fd8a96dfa9bc34f 100644
--- a/demo/finetune_sentiment_analysis.py
+++ b/demo/finetune_sentiment_analysis.py
@@ -126,7 +126,7 @@ if not args.eval:
         lr_scheduler,
         parameters=model.parameters(),
         weight_decay=args.wd,
-        apply_decay_param_fun=lambda n: param_name_to_exclue_from_weight_decay.match(n),
+        apply_decay_param_fun=lambda n: not param_name_to_exclue_from_weight_decay.match(n),
         grad_clip=g_clip)
     scaler = P.amp.GradScaler(enable=args.use_amp)
     with LogWriter(logdir=str(create_if_not_exists(args.save_dir /