diff --git a/deepspeech/exps/deepspeech2/model.py b/deepspeech/exps/deepspeech2/model.py
index bbce273f0..e00439a02 100644
--- a/deepspeech/exps/deepspeech2/model.py
+++ b/deepspeech/exps/deepspeech2/model.py
@@ -146,7 +146,7 @@ class DeepSpeech2Trainer(Trainer):
             learning_rate=config.training.lr,
             gamma=config.training.lr_decay,
             verbose=True)
-        optimizer = paddle.optimizer.SGD(  #Adam
+        optimizer = paddle.optimizer.Adam(
             learning_rate=lr_scheduler,
             parameters=model.parameters(),
             weight_decay=paddle.regularizer.L2Decay(
diff --git a/deepspeech/models/ds2_online/conv.py b/deepspeech/models/ds2_online/conv.py
index a98786e64..4a6fd5abd 100644
--- a/deepspeech/models/ds2_online/conv.py
+++ b/deepspeech/models/ds2_online/conv.py
@@ -30,6 +30,4 @@ class Conv2dSubsampling4Online(Conv2dSubsampling4):
         #b, c, t, f = paddle.shape(x) #not work under jit
         x = x.transpose([0, 2, 1, 3]).reshape([0, 0, -1])
         x_len = ((x_len - 1) // 2 - 1) // 2
-        x_len = paddle.where(x_len >= 0, x_len,
-                             paddle.zeros_like(x_len.shape, "int64"))
         return x, x_len