Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 11, 2020

• Split the graph building and running logic in

  Show description for ef24f11
  

  aaroey

  authored and

  copybara-github

  committed
  ef24f11

  Copy full SHA for ef24f11

• Internal quantization changes.

  Show description for e8ee2f7
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e8ee2f7

  Copy full SHA for e8ee2f7

• Gate tpu_embedding_summaries with an if add_summary conditional.

  Show description for 963b6e9
  

  lingvo-bot

  authored and

  copybara-github

  committed
  963b6e9

  Copy full SHA for 963b6e9

Commits on Nov 10, 2020

• Supports initialization of stateful variable for dtype=tf.int8.

  Show description for 69044d4
  

  lingvo-bot

  authored and

  copybara-github

  committed
  69044d4

  Copy full SHA for 69044d4

• Using tf.embedding_lookup with tf.tpu.cross_replica_sum will cause dimension undetermined when compile.

  Show description for a565bae
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a565bae

  Copy full SHA for a565bae

• Adds MaybeResetStepSeedFromScope.

  Show description for 11a0080
  

  lingvo-bot

  authored and

  copybara-github

  committed
  11a0080

  Copy full SHA for 11a0080

• Make ComputeLoss impls not return NestedMaps.

  Show description for d96f20f
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  d96f20f

  Copy full SHA for d96f20f

Commits on Nov 9, 2020

• Refactor gshard_builder to annotate variables directly.

  Show description for a4374e2
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a4374e2

  Copy full SHA for a4374e2

• Add quantization domain for the weights for projection and softmax layer.

  Show description for 4539de5
  

  lingvo-bot

  authored and

  copybara-github

  committed
  4539de5

  Copy full SHA for 4539de5

• Remove the dropout classes.

  Show description for 824e92b
  

  lingvo-bot

  authored and

  copybara-github

  committed
  824e92b

  Copy full SHA for 824e92b

• Add fprop_dtype_override device option to InferenceGraphExporter

  Show description for 00c654e
  

  protoget

  authored and

  copybara-github

  committed
  00c654e

  Copy full SHA for 00c654e

• Update init_from_checkpoint_rules doc.

  Show description for 900b3b1
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  900b3b1

  Copy full SHA for 900b3b1

• Make some functions eager-compatible.

  Show description for 9d7eed7
  

  lingvo-bot

  authored and

  copybara-github

  committed
  9d7eed7

  Copy full SHA for 9d7eed7

Commits on Nov 8, 2020

• Internal change

  Show description for efb1adc
  

  protoget

  authored and

  copybara-github

  committed
  efb1adc

  Copy full SHA for efb1adc

• Combine qkv matmul.

  Show description for 70b64c7
  

  lingvo-bot

  authored and

  copybara-github

  committed
  70b64c7

  Copy full SHA for 70b64c7

• Makes TransformerBatchMajorDecoder only transpose 2-D tensors in 'tgt'. When running on CPUs, the 'tgt' map contains 'transcripts', a 1-D string tensor.

  Show description for 5a6cadf
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  5a6cadf

  Copy full SHA for 5a6cadf

Commits on Nov 6, 2020

• Fix incorrect comment

  Show description for 92fc8ce
  

  lingvo-bot

  authored and

  copybara-github

  committed
  92fc8ce

  Copy full SHA for 92fc8ce

• Add limited support for indexed keys in Set.

  Show description for 01a4ddc
  

  Sergey Kishchenko

  authored and

  copybara-github

  committed
  01a4ddc

  Copy full SHA for 01a4ddc

• Automated rollback of commit 4d8aea9a1bf2a7eb7d8b8b51085ab34fb54d1345

  Show description for 475f19a
  

  bignamehyp

  authored and

  copybara-github

  committed
  475f19a

  Copy full SHA for 475f19a

• Make sure that HybridV2 works with packed input.

  Show description for c1df5b9
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c1df5b9

  Copy full SHA for c1df5b9

Commits on Nov 5, 2020

• Adds GradientAggregationOptimizer

  Show description for 4d8aea9
  

  bignamehyp

  authored and

  copybara-github

  committed
  4d8aea9

  Copy full SHA for 4d8aea9

• Add a test for nested GenericInput op.

  Show description for 374d060
  

  aaroey

  authored and

  copybara-github

  committed
  374d060

  Copy full SHA for 374d060

• Internal quantization changes.

  Show description for 09c46ab
  

  lingvo-bot

  authored and

  copybara-github

  committed
  09c46ab

  Copy full SHA for 09c46ab

• Pins global step to device cpu:0 if flag pin_vars_to_cpu is set

  Show description for e6ccbf6
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e6ccbf6

  Copy full SHA for e6ccbf6

• Makes MaybeCastToFPropDtype() work for dtypes such as tf.float32_ref.

  Show description for 7c84478
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  7c84478

  Copy full SHA for 7c84478

Commits on Nov 4, 2020

• Remove global_step arg from GenerateStepSeedPair and PostTrainingStepUpdate.

  Show description for 36498d3
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  36498d3

  Copy full SHA for 36498d3

• Remove passing of step into schedule.Value(). It always uses

  Show description for c2d520c
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  c2d520c

  Copy full SHA for c2d520c

Commits on Nov 3, 2020

• Gshard in ProjectionLayer and FeedforwardNet

  Show description for df082be
  

  protoget

  authored and

  copybara-github

  committed
  df082be

  Copy full SHA for df082be

• Add textual echo cancellation paper to publications

  Show description for 3c7d35f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3c7d35f

  Copy full SHA for 3c7d35f

Commits on Nov 2, 2020

• Remove scaling for (not p.shared_embedding_and_softmax_weights)

  Show description for fb5d5c0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  fb5d5c0

  Copy full SHA for fb5d5c0

• Use circular buffer for recurrent states of LocalSelfAttention

  Show description for 793dd10
  

  protoget

  authored and

  copybara-github

  committed
  793dd10

  Copy full SHA for 793dd10

Commits on Nov 1, 2020

• Performance optimization for conformer streaming inference

  Show description for bce8b7e
  

  protoget

  authored and

  copybara-github

  committed
  bce8b7e

  Copy full SHA for bce8b7e

• Minor changes in the debug prints.

  Show description for fc8fbe0
  

  bignamehyp

  authored and

  copybara-github

  committed
  fc8fbe0

  Copy full SHA for fc8fbe0

• Consolidate AddGlobalVN and AddPerStepVN into AddVN.

  Show description for 699757c
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  699757c

  Copy full SHA for 699757c

• Conformer inference tpu optimization

  Show description for cf64b8a
  

  protoget

  authored and

  copybara-github

  committed
  cf64b8a

  Copy full SHA for cf64b8a

Pagination

Previous

Next