diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7/train.py b/egs/librispeech/ASR/pruned_transducer_stateless7/train.py
index 300f0f0f8..20d126e07 100755
--- a/egs/librispeech/ASR/pruned_transducer_stateless7/train.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless7/train.py
@@ -66,7 +66,7 @@ from lhotse.cut import Cut
 from lhotse.dataset.sampling.base import CutSampler
 from lhotse.utils import fix_random_seed
 from model import Transducer
-from optim import Eden, NeutralGradient
+from optim import Eden, PrAdam
 from torch import Tensor
 from torch.cuda.amp import GradScaler
 from torch.nn.parallel import DistributedDataParallel as DDP
@@ -926,9 +926,8 @@ def run(rank, world_size, args):
         logging.info("Using DDP")
         model = DDP(model, device_ids=[rank])
 
-    optimizer = NeutralGradient(model.parameters(),
-                                lr=params.initial_lr,
-                                lr_for_speedup=params.initial_lr)
+    optimizer = PrAdam(model.parameters(),
+                       lr=params.initial_lr)
 
     scheduler = Eden(optimizer, params.lr_batches, params.lr_epochs)