Typo reported by https://github.com/tensorflow/models/pull/9672

PiperOrigin-RevId: 362836016

official/nlp/optimization.py

@@ -113,7 +113,7 @@ class AdamWeightDecay(tf.keras.optimizers.Adam):
   correct way of using L2 regularization/weight decay with Adam, since that will
   interact with the m and v parameters in strange ways.
 
-  Instead we want ot decay the weights in a manner that doesn't interact with
+  Instead we want to decay the weights in a manner that doesn't interact with
   the m/v parameters. This is equivalent to adding the square of the weights to
   the loss with plain (non-momentum) SGD.
   """