diff --git a/deepspeech/models/ds2/conv.py b/deepspeech/models/ds2/conv.py
index ce962a44..9548af0a 100644
--- a/deepspeech/models/ds2/conv.py
+++ b/deepspeech/models/ds2/conv.py
@@ -106,11 +106,9 @@ class ConvBn(nn.Layer):
         # reset padding part to 0
         masks = make_non_pad_mask(x_len)  #[B, T]
         masks = masks.unsqueeze(1).unsqueeze(1)  # [B, 1, 1, T]
-        # TODO(Hui Zhang): not support bool multiply
-        # masks = masks.type_as(x)
-        masks = masks.astype(x.dtype)
-        x = x.multiply(masks)
-
+        # https://github.com/PaddlePaddle/Paddle/pull/29265
+        # rhs will type promote to lhs
+        x = x * masks
         return x, x_len
 
 
diff --git a/deepspeech/models/ds2/rnn.py b/deepspeech/models/ds2/rnn.py
index 3ff91d0a..3fc52a37 100644
--- a/deepspeech/models/ds2/rnn.py
+++ b/deepspeech/models/ds2/rnn.py
@@ -308,7 +308,8 @@ class RNNStack(nn.Layer):
             x, x_len = rnn(x, x_len)
             masks = make_non_pad_mask(x_len)  #[B, T]
             masks = masks.unsqueeze(-1)  # [B, T, 1]
-            # TODO(Hui Zhang): not support bool multiply
-            masks = masks.astype(x.dtype)
-            x = x.multiply(masks)
+            # https://github.com/PaddlePaddle/Paddle/pull/29265
+            # rhs will type promote to lhs
+            x = x * masks
+
         return x, x_len
diff --git a/deepspeech/modules/conv.py b/deepspeech/modules/conv.py
index 8bf48b2c..22a16880 100644
--- a/deepspeech/modules/conv.py
+++ b/deepspeech/modules/conv.py
@@ -113,11 +113,9 @@ class ConvBn(nn.Layer):
         # reset padding part to 0
         masks = make_non_pad_mask(x_len)  #[B, T]
         masks = masks.unsqueeze(1).unsqueeze(1)  # [B, 1, 1, T]
-        # TODO(Hui Zhang): not support bool multiply
-        # masks = masks.type_as(x)
-        masks = masks.astype(x.dtype)
-        x = x.multiply(masks)
-
+        # https://github.com/PaddlePaddle/Paddle/pull/29265
+        # rhs will type promote to lhs
+        x = x * masks
         return x, x_len
 
 
diff --git a/deepspeech/modules/loss.py b/deepspeech/modules/loss.py
index f692a818..aff4854d 100644
--- a/deepspeech/modules/loss.py
+++ b/deepspeech/modules/loss.py
@@ -46,7 +46,6 @@ class CTCLoss(nn.Layer):
         # warp-ctc need activation with shape [T, B, V + 1]
         # logits: (B, L, D) -> (L, B, D)
         logits = logits.transpose([1, 0, 2])
-        # (TODO:Hui Zhang) ctc loss does not support int64 labels
         ys_pad = ys_pad.astype(paddle.int32)
         loss = self.loss(
             logits, ys_pad, hlens, ys_lens, norm_by_times=self.batch_average)
diff --git a/deepspeech/modules/rnn.py b/deepspeech/modules/rnn.py
index 0d8c9fd2..8f8b2a18 100644
--- a/deepspeech/modules/rnn.py
+++ b/deepspeech/modules/rnn.py
@@ -308,7 +308,7 @@ class RNNStack(nn.Layer):
             x, x_len = rnn(x, x_len)
             masks = make_non_pad_mask(x_len)  #[B, T]
             masks = masks.unsqueeze(-1)  # [B, T, 1]
-            # TODO(Hui Zhang): not support bool multiply
-            masks = masks.astype(x.dtype)
-            x = x.multiply(masks)
+            # https://github.com/PaddlePaddle/Paddle/pull/29265
+            # rhs will type promote to lhs
+            x = x * masks
         return x, x_len