NeuralGPU

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)

The model can be trained on the following algorithmic tasks:

• sort - Sort a symbol list

• kvsort - Sort symbol keys in dictionary

• id - Return the same symbol list

• rev - Reverse a symbol list

• rev2 - Reverse a symbol dictionary by key

• incr - Add one to a symbol value

• add - Long decimal addition

• left - First symbol in list

• right - Last symbol in list

• left-shift - Left shift a symbol list

• right-shift - Right shift a symbol list

• bmul - Long binary multiplication

• mul - Long decimal multiplication

• dup - Duplicate a symbol list with padding

• badd - Long binary addition

• qadd - Long quaternary addition

• search - Search for symbol key in dictionary

It can also be trained on the WMT English-French translation task:

• wmt - WMT English-French translation (data will be downloaded)

The value range for symbols are defined by the vocab_size flag. In particular, the values are in the range vocab_size - 1. So if you set --vocab_size=16 (the default) then --problem=rev will be reversing lists of 15 symbols, and --problem=id will be identity on a list of up to 15 symbols.

To train the model on the binary multiplication task run:

python neural_gpu_trainer.py --problem=bmul

This trains the Extended Neural GPU, to train the original model run:

python neural_gpu_trainer.py --problem=bmul --beam_size=0

While training, interim / checkpoint model parameters will be written to /tmp/neural_gpu/.

Once the amount of error gets down to what you're comfortable with, hit Ctrl-C to stop the training process. The latest model parameters will be in /tmp/neural_gpu/neural_gpu.ckpt-<step> and used on any subsequent run.

To evaluate a trained model on how well it decodes run:

python neural_gpu_trainer.py --problem=bmul --mode=1

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)

链接：https://github.com/tensorflow/models/tree/master/research/neural_gpu