Merge pull request #960 from lukaszkaiser/ngpu-corrections

Corrections and explanations for the updated Neural GPU model.

neural_gpu/README.md

 # NeuralGPU
-Code for the Neural GPU model as described
-in [[http://arxiv.org/abs/1511.08228]].
+Code for the Neural GPU model described in [[http://arxiv.org/abs/1511.08228]].
+The extended version was described in [[https://arxiv.org/abs/1610.08613]].
 
 Requirements:
 * TensorFlow (see tensorflow.org for how to install)
@@ -68,4 +68,16 @@ To interact with a model (experimental, see code) run:
 python neural_gpu_trainer.py --problem=bmul --mode=2
 ```
 
+To train on WMT data, set a larger --nmaps and --vocab_size and avoid curriculum:
+
+```
+python neural_gpu_trainer.py --problem=wmt --vocab_size=32768 --nmaps=256
+  --vec_size=256 --curriculum_seq=1.0 --max_length=60 --data_dir ~/wmt
+```
+
+With less memory, try lower batch size, e.g. `--batch_size=4`. With more GPUs
+in your system, there will be a batch on every GPU so you can run larger models.
+For example, `--batch_size=4 --num_gpus=4 --nmaps=512 --vec_size=512` will
+run a large model (512-size) on 4 GPUs, with effective batches of 4*4=16.
+
 Maintained by Lukasz Kaiser (lukaszkaiser)

neural_gpu/neural_gpu.py

@@ -128,7 +128,7 @@ def conv_gru(inpts, mem, kw, kh, nmaps, rate, cutoff, prefix, do_layer_norm,
     reset = sigmoid_cutoff(conv_lin(inpts + [mem], "r", 1.0), cutoff)
     gate = sigmoid_cutoff(conv_lin(inpts + [mem], "g", 1.0), cutoff)
   if cutoff > 10:
-    candidate = tf.tanh_hard(conv_lin(inpts + [reset * mem], "c", 0.0))
+    candidate = tanh_hard(conv_lin(inpts + [reset * mem], "c", 0.0))
   else:
     # candidate = tanh_cutoff(conv_lin(inpts + [reset * mem], "c", 0.0), cutoff)
     candidate = tf.tanh(conv_lin(inpts + [reset * mem], "c", 0.0))
@@ -273,7 +273,7 @@ class NeuralGPU(object):
 
     if backward:
       adam_lr = 0.005 * self.lr
-      adam = tf.train.AdamOptimizer(adam_lr, epsilon=2e-4)
+      adam = tf.train.AdamOptimizer(adam_lr, epsilon=1e-3)
 
       def adam_update(grads):
         return adam.apply_gradients(zip(grads, tf.trainable_variables()),

neural_gpu/neural_gpu_trainer.py

@@ -35,7 +35,7 @@ tf.app.flags.DEFINE_float("max_grad_norm", 4.0, "Clip gradients to this norm.")
 tf.app.flags.DEFINE_float("cutoff", 1.2, "Cutoff at the gates.")
 tf.app.flags.DEFINE_float("curriculum_ppx", 9.9, "Move curriculum if ppl < X.")
 tf.app.flags.DEFINE_float("curriculum_seq", 0.3, "Move curriculum if seq < X.")
-tf.app.flags.DEFINE_float("dropout", 0.0, "Dropout that much.")
+tf.app.flags.DEFINE_float("dropout", 0.1, "Dropout that much.")
 tf.app.flags.DEFINE_float("grad_noise_scale", 0.0, "Gradient noise scale.")
 tf.app.flags.DEFINE_float("max_sampling_rate", 0.1, "Maximal sampling rate.")
 tf.app.flags.DEFINE_float("length_norm", 0.0, "Length normalization.")
@@ -263,7 +263,8 @@ def initialize(sess=None):
     data.rev_vocab = rev_fr_vocab
     data.print_out("Reading development and training data (limit: %d)."
                    % FLAGS.max_train_data_size)
-    dev_set = read_data(en_dev, fr_dev, data.bins)
+    dev_set = {}
+    dev_set["wmt"] = read_data(en_dev, fr_dev, data.bins)
     def data_read(size, print_out):
       read_data_into_global(en_train, fr_train, data.bins, size, print_out)
     data_read(50000, False)
@@ -330,6 +331,7 @@ def initialize(sess=None):
   ngpu.CHOOSE_K = FLAGS.soft_mem_size
   do_beam_model = FLAGS.train_beam_freq > 0.0001 and FLAGS.beam_size > 1
   beam_size = FLAGS.beam_size if FLAGS.mode > 0 and not do_beam_model else 1
+  beam_size = min(beam_size, FLAGS.beam_size)
   beam_model = None
   def make_ngpu(cur_beam_size, back):
     return ngpu.NeuralGPU(