Skip to content

Releases: OpenNMT/OpenNMT-tf

OpenNMT-tf 2.32.0

04 Aug 08:42
Compare
Choose a tag to compare

New features

  • Support TensorFlow 2.12 and 2.13
  • Make timeout value configurable while searching for an optimal batch size

OpenNMT-tf 2.31.0

13 Jan 15:54
Compare
Choose a tag to compare

New features

  • Add option --jit_compile to compile the model with XLA (only applied in training at the moment)

Fixes and improvements

  • Improve correctness of gradient accumulation and multi-GPU training by normalizing the gradients with the true global batch size instead of using an approximation
  • Report the total number of tokens per second in the training logs, in addition to the source and target numbers
  • Relax the sacreBLEU version requirement to include any 2.x versions

OpenNMT-tf 2.30.0

12 Dec 13:16
Compare
Choose a tag to compare

Changes

  • The model attribute ctranslate2_spec has been removed as it is no longer relevant with the new CTranslate2 converter
  • The global gradient norm is no longer reported in TensorBoard because it was misleading: it did not take into account gradient accumulation and multi-GPU

New features

  • Support TensorFlow 2.11 (note that the new Keras optimizers are not yet supported, if you are creating optimizers manually please use an optimizer in tf.keras.optimizers.legacy for now)
  • Support CTranslate2 3.0
  • Add training parameter pad_to_bucket_boundary to pad the batch length to a multiple of length_bucket_width (this is useful to reduce the number of recompilation with XLA)
  • Integrate the scorers chrf and chrf++ from SacreBLEU

Fixes and improvements

  • Fix error when training with Horovod and using an early stopping condition
  • Fix error when using guided alignment with mixed precision

OpenNMT-tf 2.29.1

03 Oct 09:29
Compare
Choose a tag to compare

Fixes and improvements

  • Fix error when using gzipped training data files
  • Remove unnecessary casting in MultiHeadAttention for a small performance improvement

OpenNMT-tf 2.29.0

26 Sep 15:11
Compare
Choose a tag to compare

New features

  • Support TensorFlow 2.10
  • Add model configurations ScalingNmtEnDe and ScalingNmtEnFr from Ott et al. 2018
  • Add embedding parameter EmbeddingsSharingLevel.AUTO to automatically share embeddings when the vocabulary is shared
  • Extend method Runner.average_checkpoints to accept a list of checkpoints to average

Fixes and improvements

  • Make batch size autotuning faster when using gradient accumulation

OpenNMT-tf 2.28.0

29 Jul 09:28
Compare
Choose a tag to compare

New features

  • Add initial_learning_rate parameter to the InvSqrtDecay schedule
  • Add new arguments to the Transformer constructor:
    • mha_bias: to disable bias terms in the multi-head attention (as presented in the original paper)
    • output_layer_bias: to disable bias in the output linear layer

Fixes and improvements

  • Fix incorrect dtype for SequenceRecordInputter length vector
  • Fix rounding error when batching datasets which could make the number of tokens in a batch greater than the configured batch size
  • Fix deprecation warning when using distutils.version.LooseVersion, use packaging.version.Version instead
  • Make the length dimension unknown in the dataset used for batch size autotuning so that it matches the behavior in training
  • Update SacreBLEU requirement to include new version 2.2

OpenNMT-tf 2.27.1

02 Jun 12:53
Compare
Choose a tag to compare

Fixes and improvements

  • Fix evaluation and scoring with language models

OpenNMT-tf 2.27.0

30 May 09:23
Compare
Choose a tag to compare

Changes

  • Remove support for older TensorFlow versions 2.4 and 2.5
  • Remove support for deprecated Python version 3.6

New features

  • Support TensorFlow 2.9
  • Integrate the new CTranslate2 converter to export more Transformer variants, including multi-features models

Fixes and improvements

  • Fix error when loading the SavedModel of Transformer models with relative position representations
  • Fix dataset error in inference with language models
  • Fix batch size autotuning error with language models
  • Fix division by zero error on some systems when the time to the last training log is too small

OpenNMT-tf 2.26.1

31 Mar 10:54
Compare
Choose a tag to compare

Fixes and improvements

  • Fix documentation build error

OpenNMT-tf 2.26.0

31 Mar 10:54
Compare
Choose a tag to compare

New features

  • Add learning rate schedule InvSqrtDecay
  • Enable CTranslate2 conversion for models using GELU or Swish activations

Fixes and improvements

  • Fix inference error when using the decoding_noise parameter
  • Clarify the inference log about buffered predictions