diff --git a/seq2seq/models/seq2seq_model.py b/seq2seq/models/seq2seq_model.py
index 423ffb75..2825c24c 100644
--- a/seq2seq/models/seq2seq_model.py
+++ b/seq2seq/models/seq2seq_model.py
@@ -74,7 +74,7 @@ def _clip_gradients(self, grads_and_vars):
     clipped_gradients = []
     variables = []
     for gradient, variable in grads_and_vars:
-      if "embedding" in variable.name:
+      if "embedding" in variable.name and gradient != None:
         tmp = tf.clip_by_norm(
             gradient.values, self.params["optimizer.clip_embed_gradients"])
         gradient = tf.IndexedSlices(tmp, gradient.indices, gradient.dense_shape)