glample / rnn-benchmarks Goto Github PK

Can you add caffe rnn model to test?

You should consider this one, AFAIK it is much more faster than Element Research. https://github.com/jcjohnson/torch-rnn

Thank you for your RNN benchmarking work! Hidden size is somewhat arbitrary, so there is virtually no reason (other than being really tight on memory) to make it, say, 500 instead of 512. Underlying gemms in RNNs/LSTM's usually have better efficiency with power-of-two sizes. It would be good to have benchmarks for hidden sizes 128, 512, 1024. Also, pretty often people use bigger mini-batches than 20, so benchmarking batches of 32/64 also make sense.

Great benchmark for evaluting RNN units with different frameworks. But the world is changing fast and TensorFlow has evolved to version 1.4.0rc0. The benchmark for TenforFlow depends on some deprecated APIs which have been removed or replaced.

$ python rnn.py -n basic_lstm -b 32 -l 128 -s 30
basic_lstm
Compiling...
Traceback (most recent call last):
  File "rnn.py", line 59, in <module>
    output, _cell_state = rnn.rnn(cell, x, dtype=tf.float32)
AttributeError: 'module' object has no attribute 'rnn'

Hope you can update the code to follow the latest features of those frameworks. Thanks for your great works.

You need to add an open-source LICENSE.

I recommend BSD or MIT.

Add your and your organization’s name to the copyright notice in the LICENSE.txt.

I cannot do this for you as it must be done with your commit.

Also please send me an email so I know yours.

Is there any RNN and LSTM benchmark available for android? Any help regarding this will highly be appreciated.

Since the libraries are very actively maintained and improved upon, can you add versions of the libraries that are used for the comparisons? (Unless I'm somehow missing this information on the README page).