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

v1.6.1#751

Merged

lukaszkaiser merged 23 commits intotensorflow:masterfrom

rsepassi:push

Apr 26, 2018

Commits on Apr 26, 2018

Make base Transformer behave well on 2D data for baselines.

Lukasz Kaiser
authored and
Ryan Sepassi
committed
Shuffle in parallel
Ryan Sepassi
committed
Only make shuffle parallel if >1 file
Ryan Sepassi
committed
Create data_dir before vocab save
Ryan Sepassi
committed
Recurrent Transformers.

T2T Team
authored and
Ryan Sepassi
committed
Use TF 1.7 in Cloud TPU path, use gcloud beta, add --cloud_skip_confirmation
Ryan Sepassi
committed
Update ImportError logging in data_generators.all_problems
Ryan Sepassi
committed
faster implementation of relative attention for masked-self-attention case.

nshazeer
authored and
Ryan Sepassi
committed
don't crash if hparams.max_length is unspecified

nshazeer
authored and
Ryan Sepassi
committed
Add support for PREDICT on TPU, tested with resnet50
Ryan Sepassi
committed
Version bump
Ryan Sepassi
committed
Implement span-positioned SquAD text problem.

T2T Team
authored and
Ryan Sepassi
committed
Release Wikipedia Summarization dataset
Ryan Sepassi
committed
Fix pending bug :(

Niki Parmar
authored and
Ryan Sepassi
committed
Internal merges
Ryan Sepassi
committed
don't crash if hparams.max_length is unspecified

T2T Team
authored and
Ryan Sepassi
committed
StringIO Py2/3 fix
Ryan Sepassi
committed
Internal merges
Ryan Sepassi
committed
Sort the losses_dict before passing it to the sum op, in order to make the tf graph deterministic.

T2T Team
authored and
Ryan Sepassi
committed
Corrections to model-based RL code.

Lukasz Kaiser
authored and
Ryan Sepassi
committed
Add h5py to deps
Ryan Sepassi
committed
Autoencoder improvements and related corrections; ordered and discrete residual MNIST samples look reasonable.

Lukasz Kaiser
authored and
Ryan Sepassi
committed
Add travis ignore for r_transformer_test
Ryan Sepassi
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

v1.6.1#751

v1.6.1#751
lukaszkaiser merged 23 commits intotensorflow:masterfrom
rsepassi:push

Commits on Apr 26, 2018

Make base Transformer behave well on 2D data for baselines.

Shuffle in parallel

Only make shuffle parallel if >1 file

Create data_dir before vocab save

Recurrent Transformers.

Use TF 1.7 in Cloud TPU path, use gcloud beta, add --cloud_skip_confirmation

Update ImportError logging in data_generators.all_problems

faster implementation of relative attention for masked-self-attention case.

don't crash if hparams.max_length is unspecified

Add support for PREDICT on TPU, tested with resnet50

Version bump

Implement span-positioned SquAD text problem.

Release Wikipedia Summarization dataset

Fix pending bug :(

Internal merges

don't crash if hparams.max_length is unspecified

StringIO Py2/3 fix

Internal merges

Sort the losses_dict before passing it to the sum op, in order to make the tf graph deterministic.

Corrections to model-based RL code.

Add h5py to deps

Autoencoder improvements and related corrections; ordered and discrete residual MNIST samples look reasonable.

Add travis ignore for r_transformer_test