Commits

Branch selector

 master

User selector

lingvo-bot

Datepicker

All time

Commit History

Commits on Aug 4, 2020

• Implements the routing attention based on k-means clustering.

  Show description for 11a216c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  11a216c

  Copy full SHA for 11a216c

• Add generic_input op unit test for usage with tf.data

  Show description for 68dfe3a
  

  lingvo-bot

  authored and

  copybara-github

  committed
  68dfe3a

  Copy full SHA for 68dfe3a

Commits on Aug 3, 2020

• base_model.py refactor.

  Show description for 8ff6a9f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  8ff6a9f

  Copy full SHA for 8ff6a9f

Commits on Aug 1, 2020

• in moe_layers.py allow state layer updates

  Show description for b4fd694
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b4fd694

  Copy full SHA for b4fd694

• Add autoregressive decoding support for LocalSelfAttention.

  Show description for b4e6e49
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b4e6e49

  Copy full SHA for b4e6e49

• Add per_core_infeed option to allow handling of per host infeed with data parallelism while num_partitions is > 1.

  Show description for 391780f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  391780f

  Copy full SHA for 391780f

Commits on Jul 31, 2020

• Support streaming ground truth bounding boxes to reduce memory consumption for GroundTruthAugmentor preprocessor.

  Show description for e07e8d9
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e07e8d9

  Copy full SHA for e07e8d9

• Add a flag for softmax scaling code motion and set it to false by default.

  Show description for e876b48
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e876b48

  Copy full SHA for e876b48

Commits on Jul 30, 2020

• Allow fprop_dtype propagation to ShardedVarLayer.

  Show description for 87f7963
  

  lingvo-bot

  authored and

  copybara-github

  committed
  87f7963

  Copy full SHA for 87f7963

• Add v2_padding option to conv_layers_builder.py

  Show description for 0f13467
  

  lingvo-bot

  authored and

  copybara-github

  committed
  0f13467

  Copy full SHA for 0f13467

• Allow IdentityLayer inputs to be NestedMap.

  Show description for cf0cb4d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  cf0cb4d

  Copy full SHA for cf0cb4d

• Remove AddTrainerVerboseTensor.

  Show description for 5d255ea
  

  lingvo-bot

  authored and

  copybara-github

  committed
  5d255ea

  Copy full SHA for 5d255ea

• Rename relative_atten_util to attention_util

  Show description for 0dc242d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  0dc242d

  Copy full SHA for 0dc242d

• Computes SmoothL1 loss on the coordinates of each corner.

  Show description for a56d76e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a56d76e

  Copy full SHA for a56d76e

• Implement the k-means clustering of the routing transformer.

  Show description for 3e9cc1c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3e9cc1c

  Copy full SHA for 3e9cc1c

Commits on Jul 29, 2020

• Allow objects created with InstantiableParams to accept extra constructor args.

  Show description for 33cbbfa
  

  lingvo-bot

  authored and

  copybara-github

  committed
  33cbbfa

  Copy full SHA for 33cbbfa

• Adds has_bias option for FeedForwardNet.

  Show description for 2d1fa86
  

  lingvo-bot

  authored and

  copybara-github

  committed
  2d1fa86

  Copy full SHA for 2d1fa86

Commits on Jul 28, 2020

• Fix CategoricalLayerNorm to use theta instead of self.vars during FProp.

  Show description for a38fa5a
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a38fa5a

  Copy full SHA for a38fa5a

• Skip None tensors when converting subgraphs dictionary to proto

  Show description for fcc308e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  fcc308e

  Copy full SHA for fcc308e

• test BatchNormLayer with UpdateBatchNormVars

  Show description for c80282d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c80282d

  Copy full SHA for c80282d

Commits on Jul 27, 2020

• Add xla_sharding annotation to ff and attention vars of batch major attention.

  Show description for 4868d48
  

  lingvo-bot

  authored and

  copybara-github

  committed
  4868d48

  Copy full SHA for 4868d48

• Create separate variables for scales and biases in Categorical layer norm.

  Show description for 394729f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  394729f

  Copy full SHA for 394729f

Commits on Jul 24, 2020

• Add mass_task_ids for TextPackedInputV2

  Show description for ca77e50
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ca77e50

  Copy full SHA for ca77e50

Commits on Jul 23, 2020

• Split MatplotlibFigureSummary.Finalize into a FinalizeImage function that simply returns the uint8 image tensor and Finalize (which creates the TF1 image summary).

  Show description for ffa932b
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ffa932b

  Copy full SHA for ffa932b

Commits on Jul 20, 2020

• Enables the target_sequence_sampler to use a stop_fn that early terminates the Recurrent loop when all sequences in the batch end in target_eos_id.

  Show description for 6565218
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6565218

  Copy full SHA for 6565218

Commits on Jul 17, 2020

• Add support to insert adapter layers after each lstm layer in the decoder.

  Show description for f7d4cb6
  

  lingvo-bot

  authored and

  copybara-github

  committed
  f7d4cb6

  Copy full SHA for f7d4cb6

Commits on Jul 15, 2020

• Implements NormalizedLSTMCellSimple allowing LSTMCell to customize

  Show description for 719a65f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  719a65f

  Copy full SHA for 719a65f

Commits on Jul 14, 2020

• Fix broken convolutions with strides and/or VALID padding.

  Show description for 621d648
  

  lingvo-bot

  authored and

  copybara-github

  committed
  621d648

  Copy full SHA for 621d648

• Move scalings from the side of T^2 to the side of T for better performance.

  Show description for 75a8556
  

  lingvo-bot

  authored and

  copybara-github

  committed
  75a8556

  Copy full SHA for 75a8556

Commits on Jul 13, 2020

• Fix shape in SampleSequenceDecode

  Show description for ad554e3
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ad554e3

  Copy full SHA for ad554e3

Commits on Jul 10, 2020

• Write inference graphs for all tasks of MultiTaskModels.

  Show description for 1a56917
  

  lingvo-bot

  authored and

  copybara-github

  committed
  1a56917

  Copy full SHA for 1a56917

• Adds add_image_summary option for classifier model.

  Show description for e3f4334
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e3f4334

  Copy full SHA for e3f4334

Commits on Jul 9, 2020

• Increase default tpu_steps_per_loop to 1000 for better TPU fleet efficiency.

  Show description for ad6d520
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ad6d520

  Copy full SHA for ad6d520

• Keep Softmax in F32 to fix the convergence issue when fprop_dtype is bf16

  Show description for d9312dd
  

  lingvo-bot

  authored and

  copybara-github

  committed
  d9312dd

  Copy full SHA for d9312dd

Commits on Jul 7, 2020

• Support dataclasses in hyperparams.

  Show description for 6491f4e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6491f4e

  Copy full SHA for 6491f4e

Pagination

Previous

Next