Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Oct 13, 2018

• internal merge of PR #1140

  Show description for 0f6f2d8
  

  koz4k

  authored and

  Copybara-Service

  committed
  0f6f2d8

  Copy full SHA for 0f6f2d8

• Implement new RL pipeline (#1140)

  Show description for e470049
  

  koz4k

  authored and

  lukaszkaiser

  committed
  e470049

  Copy full SHA for e470049

• EPVA hparams inheriting from SV2P hparams.

  Show description for 6f1ba1d
  

  T2T Team

  authored and

  Copybara-Service

  committed
  6f1ba1d

  Copy full SHA for 6f1ba1d

Commits on Oct 12, 2018

• Fix multi_problem_test.py.

  Show description for 628bf1f
  

  Dustin Tran

  authored and

  Copybara-Service

  committed
  628bf1f

  Copy full SHA for 628bf1f

• Add a check asserting that arguments to binary operations have the same dtype.

  Show description for 2e89303
  

  nshazeer

  authored and

  Copybara-Service

  committed
  2e89303

  Copy full SHA for 2e89303

• Get bf16 activations working with mixture-of-experts.

  Show description for ee89d7b
  

  nshazeer

  authored and

  Copybara-Service

  committed
  ee89d7b

  Copy full SHA for ee89d7b

• shared_rel is missing in hparams while running model=img2img_transformer and problem=img2img_celeba with hparams_set=img2img_transformer_base_tpu

  Show description for c8d7157
  

  afrozenator

  authored and

  Copybara-Service

  committed
  c8d7157

  Copy full SHA for c8d7157

• Improve reward prediction in the basic model, use constant learning rate.

  Show description for 533db5a
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  533db5a

  Copy full SHA for 533db5a

Commits on Oct 11, 2018

• internal merge of PR #1137

  Show description for b4df57e
  

  koz4k

  authored and

  Copybara-Service

  committed
  b4df57e

  Copy full SHA for b4df57e

• internal

  Show description for 7c14ecd
  

  T2T Team

  authored and

  Copybara-Service

  committed
  7c14ecd

  Copy full SHA for 7c14ecd

• Move Model-Based RL hparams to separate module. (#1137)

  

  koz4k

  authored and

  afrozenator

  committed
  1b16365

  Copy full SHA for 1b16365

• Removing a print that slipped through and using more appropriate variable names.

  Show description for be3ced5
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  be3ced5

  Copy full SHA for be3ced5

• Allow to pass Problem class as well as name in trainer_lib.

  Show description for fb92e62
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  fb92e62

  Copy full SHA for fb92e62

• Supports pretraining the Glow encoder to get a stable latent mapping before learning a conditional distribution on top.

  Show description for 75da8f1
  

  MechCoder

  authored and

  Copybara-Service

  committed
  75da8f1

  Copy full SHA for 75da8f1

• Bring linear scheduled sampling mode back for RL.

  Show description for 7da0b80
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  7da0b80

  Copy full SHA for 7da0b80

• merge pull requests.

  Show description for 06539f3
  

  nshazeer

  authored and

  Copybara-Service

  committed
  06539f3

  Copy full SHA for 06539f3

• fixes to mesh tf split (#1128)

  

  WuTheFWasThat

  authored and

  lukaszkaiser

  committed
  50fa0fb

  Copy full SHA for 50fa0fb

• fix mtf adafactor implementation for beta1>0 (#1126)

  

  WuTheFWasThat

  authored and

  lukaszkaiser

  committed
  f3b2add

  Copy full SHA for f3b2add

Commits on Oct 10, 2018

• merging all the scheduled samplings.

  Show description for 8684670
  

  mbz

  authored and

  Copybara-Service

  committed
  8684670

  Copy full SHA for 8684670

• Make a RNN for predicting latent state structure and use 16 frames grayscale default for stochastic discrete.

  Show description for a2cc2fe
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  a2cc2fe

  Copy full SHA for a2cc2fe

• internal merge of PR #1131

  Show description for 1cd6b0e
  

  konradczechowski

  authored and

  Copybara-Service

  committed
  1cd6b0e

  Copy full SHA for 1cd6b0e

• Preprocessing in t2 t gym env (#1131)

  Show description for 1bb0f56
  

  konradczechowski

  authored and

  lukaszkaiser

  committed
  1bb0f56

  Copy full SHA for 1bb0f56

• Allow decoding from the dual autoencoder and increase number of shared bits.

  Show description for 7cb5e55
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  7cb5e55

  Copy full SHA for 7cb5e55

• internal merge of PR #1124

  Show description for 53cdd2a
  

  koz4k

  authored and

  Copybara-Service

  committed
  53cdd2a

  Copy full SHA for 53cdd2a

• RL pipeline refactoring: Turn T2TEnv into a Problem and enable generating data with it (#1124)

  Show description for db631cd
  

  koz4k

  authored and

  lukaszkaiser

  committed
  db631cd

  Copy full SHA for db631cd

Commits on Oct 9, 2018

• Add instruction for the toy_model.

  Show description for 6da0ee2
  

  Youlong Cheng

  authored and

  Copybara-Service

  committed
  6da0ee2

  Copy full SHA for 6da0ee2

• Fix Python 3 compatibility issue in MeshTensorflow

  Show description for b243205
  

  Youlong Cheng

  authored and

  Copybara-Service

  committed
  b243205

  Copy full SHA for b243205

Commits on Oct 8, 2018

• internal merge of PR #1119

  Show description for 1e81346
  

  koz4k

  authored and

  Copybara-Service

  committed
  1e81346

  Copy full SHA for 1e81346

• RL pipeline refactoring: T2TGymEnv (#1119)

  Show description for c92c3a1
  

  koz4k

  authored and

  lukaszkaiser

  committed
  c92c3a1

  Copy full SHA for c92c3a1

• This fixes an important bug in the computation of video metrics.

  Show description for dd71753
  

  MechCoder

  authored and

  Copybara-Service

  committed
  dd71753

  Copy full SHA for dd71753

Commits on Oct 5, 2018

• Rm eval summary hooks (and tensor hook) in case of TPUs.

  Show description for 84cfac7
  

  Dustin Tran

  authored and

  Copybara-Service

  committed
  84cfac7

  Copy full SHA for 84cfac7

• Some changes to mixture-of-experts in mesh-tensorflow: At the inner level of hierarchical MoE, prioritize examples that were the first choice at the outer level. Change default capacity factor to…

  Show description for 3750e14
  

  nshazeer

  authored and

  Copybara-Service

  committed
  3750e14

  Copy full SHA for 3750e14

• internal

  Show description for a4f958a
  

  Ryan Sepassi

  authored and

  Copybara-Service

  committed
  a4f958a

  Copy full SHA for a4f958a

• Enable fast beam decoding from dataset on TPU.

  Show description for 59a86ee
  

  T2T Team

  authored and

  Copybara-Service

  committed
  59a86ee

  Copy full SHA for 59a86ee

• Add the option to use grayscale and correct registration for autoencoded problems; also bring tests back.

  Show description for 17f9a9a
  

  Lukasz Kaiser

  authored and

  Copybara-Service

  committed
  17f9a9a

  Copy full SHA for 17f9a9a