This repository was archived by the owner on Jul 7, 2023. It is now read-only.

v1.5.5#646

Merged

lukaszkaiser merged 25 commits intotensorflow:masterfrom

Mar 10, 2018

Commits on Mar 9, 2018

Internal change
Ryan Sepassi
committed
Set evaluation_master to master in RunConfig
Ryan Sepassi
committed
Add option to do soft EM instead of hard EM

T2T Team
authored and
Ryan Sepassi
committed
Add an inv_temp hparam for controlling softness of EM

T2T Team
authored and
Ryan Sepassi
committed
More Librispeech subsets to help with mixed clean and noisy data training and eval.

T2T Team
authored and
Ryan Sepassi
committed
proper em - P(c_i) is computed using ema_count instead of actual counts

T2T Team
authored and
Ryan Sepassi
committed
increase pseudo-count to 1.0 and now there's no NaN in training

T2T Team
authored and
Ryan Sepassi
committed
Use logits instead of probs to compute supervised attention loss.

T2T Team
authored and
Ryan Sepassi
committed
Why do we need stop gradient here?

T2T Team
authored and
Ryan Sepassi
committed
Add expected_attention_loss_type hparam to supervised_attention to allow experimentation with multiple loss functions.

T2T Team
authored and
Ryan Sepassi
committed
ema_count trainable should be False; this was causing the weird dp behavior

T2T Team
authored and
Ryan Sepassi
committed
Fix multi-logit loss computation error.

aidangomez
authored and
Ryan Sepassi
committed
Basic autoencoder and improvements in image modality.

Lukasz Kaiser
authored and
Ryan Sepassi
committed

Commits on Mar 10, 2018

Change batch size for hparam config

Niki Parmar
authored and
Ryan Sepassi
committed
Make Vanilla GAN work, based on Compare GAN code.

Lukasz Kaiser
authored and
Ryan Sepassi
committed
internal

Lukasz Kaiser
authored and
Ryan Sepassi
committed
Bump release number.

Lukasz Kaiser
authored and
Ryan Sepassi
committed
Documentation for cloud TPU for Image Transformer. Additional default tpu hparams for imagetransformer

Niki Parmar
authored and
Ryan Sepassi
committed
Add smoothed L0 prior and trainable logits for cluster probabilities.

Ashish Vaswani
authored and
Ryan Sepassi
committed
Added the ema count smoothing update inside the else.

Ashish Vaswani
authored and
Ryan Sepassi
committed
Make text_encoder unicode conversion a pass-through
Ryan Sepassi
committed
Pass in decode_hp to _interactive_input_fn and remove summaries when

T2T Team
authored and
Ryan Sepassi
committed
six.iteritems for Py3
Ryan Sepassi
committed
Update Travis tests for Py3 to run TF 1.6
Ryan Sepassi
committed
Update ISSUE_TEMPLATE
Ryan Sepassi
committed