Contents Menu Expand Light mode Dark mode Auto light/dark, in light mode Auto light/dark, in dark mode Skip to content
RETURNN documentation
RETURNN documentation

Getting Started

  • Installation
  • Basic Usage
  • RETURNN frontend
  • Technological Overview
  • Tensor and Dim
  • Recurrency
  • Training
  • Native operations
  • RETURNN as Framework
  • Recent development of RETURNN
  • TensorFlow LSTM Benchmark
  • Frequently Asked Questions

User Guide

  • Network Structure
  • Dataset Input/Output
  • Recurrent Sub-Networks
  • Extending RETURNN

Reference

  • Configuration Parameters
    • behavior_version
    • General Settings
    • Training
    • Optimizer Settings
    • Pretraining
    • Model Loading
    • Generation and Search
    • Debugging
  • Datasets
    • Generic Datasets
    • Text Datasets
    • Audio Datasets
    • Dataset Combination
  • Layers / Modules
    • Basic Layers
    • Shape and Type Modification
    • Recurrent Layers
    • Recurrent Units
    • Attention Layers
    • Normalization Layers
    • Regularization Layers
    • Custom Layers
    • Utility Layers
    • Loss Functions
    • Softmax Layers
    • Generative Layers
    • Alignment Layers
  • Optimizer

Advanced Topics

  • Pre-Training / Dynamic Networks
  • Multi GPU training with PyTorch
  • Multi GPU training with TensorFlow
  • Debugging
  • Profiling (Python generic)
  • Profiling (PyTorch)
  • Profiling (TensorFlow)
  • Deterministic training
  • Test suite

Applications

  • Speech Recognition
  • Language Modelling
  • Machine Translation

Internals

  • API
    • returnn
    • returnn.__main__
    • returnn.__old_mod_loader__
    • returnn.__setup__
    • returnn.config
    • returnn.datasets
    • returnn.datasets.audio
    • returnn.datasets.basic
    • returnn.datasets.bundle_file
    • returnn.datasets.cached
    • returnn.datasets.cached2
    • returnn.datasets.distrib_files
    • returnn.datasets.generating
    • returnn.datasets.hdf
    • returnn.datasets.huggingface
    • returnn.datasets.lm
    • returnn.datasets.map
    • returnn.datasets.meta
    • returnn.datasets.multi_proc
    • returnn.datasets.normalization_data
    • returnn.datasets.numpy_dump
    • returnn.datasets.postprocessing
    • returnn.datasets.raw_wav
    • returnn.datasets.sprint
    • returnn.datasets.stereo
    • returnn.datasets.text_dict
    • returnn.datasets.util
    • returnn.datasets.util.feature_extraction
    • returnn.datasets.util.strings
    • returnn.datasets.util.vocabulary
    • returnn.engine
    • returnn.engine.base
    • returnn.engine.batch
    • returnn.extern
    • returnn.extern.HawkAaronWarpTransducer
    • returnn.extern.WarpRna
    • returnn.extern.WarpRna.__main__
    • returnn.extern.graph_editor
    • returnn.extern.graph_editor.edit
    • returnn.extern.graph_editor.reroute
    • returnn.extern.graph_editor.select
    • returnn.extern.graph_editor.subgraph
    • returnn.extern.graph_editor.transform
    • returnn.extern.graph_editor.util
    • returnn.extern.official_tf_resnet
    • returnn.extern.official_tf_resnet.resnet_model
    • returnn.forward_iface
    • returnn.frontend
    • returnn.frontend._backend
    • returnn.frontend._cache
    • returnn.frontend._native
    • returnn.frontend._numpy_backend
    • returnn.frontend._random_journal
    • returnn.frontend._utils
    • returnn.frontend.array_
    • returnn.frontend.assert_
    • returnn.frontend.attention
    • returnn.frontend.audio
    • returnn.frontend.audio.mel
    • returnn.frontend.audio.specaugment
    • returnn.frontend.backend
    • returnn.frontend.build_from_dict
    • returnn.frontend.cond
    • returnn.frontend.const
    • returnn.frontend.container
    • returnn.frontend.control_flow_ctx
    • returnn.frontend.conv
    • returnn.frontend.conversions
    • returnn.frontend.conversions.espnet_e_branchformer
    • returnn.frontend.conversions.hf_llama
    • returnn.frontend.conversions.torch_nn
    • returnn.frontend.decoder
    • returnn.frontend.decoder.transformer
    • returnn.frontend.device
    • returnn.frontend.dims
    • returnn.frontend.dropout
    • returnn.frontend.dtype
    • returnn.frontend.encoder
    • returnn.frontend.encoder.base
    • returnn.frontend.encoder.conformer
    • returnn.frontend.encoder.conformer_v2
    • returnn.frontend.encoder.e_branchformer
    • returnn.frontend.encoder.transformer
    • returnn.frontend.gradient
    • returnn.frontend.graph
    • returnn.frontend.hooks
    • returnn.frontend.init
    • returnn.frontend.label_smoothing
    • returnn.frontend.linear
    • returnn.frontend.loop
    • returnn.frontend.loss
    • returnn.frontend.math_
    • returnn.frontend.matmul
    • returnn.frontend.module
    • returnn.frontend.nested
    • returnn.frontend.normalization
    • returnn.frontend.parameter
    • returnn.frontend.parametrizations
    • returnn.frontend.parametrize
    • returnn.frontend.piecewise_linear
    • returnn.frontend.rand
    • returnn.frontend.rec
    • returnn.frontend.reduce
    • returnn.frontend.run_ctx
    • returnn.frontend.signal
    • returnn.frontend.state
    • returnn.frontend.stepwise_scheduler
    • returnn.frontend.tensor_array
    • returnn.frontend.types
    • returnn.import_
    • returnn.import_.common
    • returnn.import_.git
    • returnn.import_.import_
    • returnn.learning_rate_control
    • returnn.log
    • returnn.native_op
    • returnn.pretrain
    • returnn.sprint
    • returnn.sprint.cache
    • returnn.sprint.control
    • returnn.sprint.error_signals
    • returnn.sprint.extern_interface
    • returnn.sprint.interface
    • returnn.tensor
    • returnn.tensor._dim_extra
    • returnn.tensor._tensor_extra
    • returnn.tensor._tensor_mixin_base
    • returnn.tensor._tensor_op_overloads
    • returnn.tensor.control_flow_ctx
    • returnn.tensor.dim
    • returnn.tensor.marked_dim
    • returnn.tensor.tensor
    • returnn.tensor.tensor_dict
    • returnn.tensor.utils
    • returnn.tf
    • returnn.tf.compat
    • returnn.tf.data_pipeline
    • returnn.tf.distributed
    • returnn.tf.engine
    • returnn.tf.frontend_layers
    • returnn.tf.frontend_layers._backend
    • returnn.tf.frontend_layers._utils
    • returnn.tf.frontend_layers.cond
    • returnn.tf.frontend_layers.config_entry_points
    • returnn.tf.frontend_layers.debug_eager_mode
    • returnn.tf.frontend_layers.dims
    • returnn.tf.frontend_layers.layer
    • returnn.tf.frontend_layers.loop
    • returnn.tf.frontend_layers.make_layer
    • returnn.tf.frontend_layers.masked_computation
    • returnn.tf.frontend_layers.parameter_assign
    • returnn.tf.frontend_layers.prev_tensor_ref
    • returnn.tf.frontend_low_level
    • returnn.tf.frontend_low_level._backend
    • returnn.tf.horovod
    • returnn.tf.hyper_param_tuning
    • returnn.tf.layers
    • returnn.tf.layers.base
    • returnn.tf.layers.basic
    • returnn.tf.layers.rec
    • returnn.tf.layers.segmental_model
    • returnn.tf.layers.signal_processing
    • returnn.tf.layers.variable
    • returnn.tf.native_op
    • returnn.tf.network
    • returnn.tf.sprint
    • returnn.tf.updater
    • returnn.tf.util
    • returnn.tf.util.basic
    • returnn.tf.util.data
    • returnn.tf.util.gradient_checkpoint
    • returnn.tf.util.ken_lm
    • returnn.tf.util.open_fst
    • returnn.torch
    • returnn.torch.data
    • returnn.torch.data.extern_data
    • returnn.torch.data.pipeline
    • returnn.torch.data.queued_data_iter
    • returnn.torch.data.returnn_dataset_wrapper
    • returnn.torch.data.tensor_utils
    • returnn.torch.distributed
    • returnn.torch.engine
    • returnn.torch.frontend
    • returnn.torch.frontend._backend
    • returnn.torch.frontend._rand
    • returnn.torch.frontend.bridge
    • returnn.torch.frontend.compile_helper
    • returnn.torch.frontend.raw_ops
    • returnn.torch.optim
    • returnn.torch.optim.lion
    • returnn.torch.updater
    • returnn.torch.util
    • returnn.torch.util.array_
    • returnn.torch.util.assert_
    • returnn.torch.util.debug_inf_nan
    • returnn.torch.util.diagnose_gpu
    • returnn.torch.util.distributed
    • returnn.torch.util.exception_helper
    • returnn.torch.util.gradient_checkpoint
    • returnn.torch.util.module
    • returnn.torch.util.native_op
    • returnn.torch.util.native_op_code_compiler
    • returnn.torch.util.rope
    • returnn.torch.util.scaled_gradient
    • returnn.util
    • returnn.util.basic
    • returnn.util.better_exchook
    • returnn.util.bpe
    • returnn.util.collect_outputs_dict
    • returnn.util.cuda_env
    • returnn.util.debug
    • returnn.util.debug_helpers
    • returnn.util.file_cache
    • returnn.util.fsa
    • returnn.util.hot_reload
    • returnn.util.literal_py_to_pickle
    • returnn.util.lru_cache
    • returnn.util.math
    • returnn.util.multi_proc_manager_with_watchdog
    • returnn.util.multi_proc_non_daemonic_spawn
    • returnn.util.native_code_compiler
    • returnn.util.pprint
    • returnn.util.py_ext_mod_compiler
    • returnn.util.result_with_reason
    • returnn.util.sig_proc
    • returnn.util.task_system
    • returnn.util.train_proc_manager
    • returnn.util.watch_memory
  • TensorFlow Beam Search
Back to top

returnn.engine¶

This package contains basic utilities for the backend engine, specifically for returnn.tf.engine.Engine or returnn.theano.engine.Engine, and also the base class returnn.engine.base.EngineBase.

Next
returnn.engine.base
Previous
returnn.datasets.util.vocabulary
Copyright © 2014–2023, RETURNN contributors
Made with Sphinx and @pradyunsg's Furo