kd

[[Source]]

import kauldron as kd

Kauldron public API.

from kauldron import kd

All symbols

Module

kd	Kauldron public API.
kd._filter_logs	Helper to filter logs from verbose modules.
kd.ckpts	Checkpoints API.
kd.ckpts.items	Checkpoint handler.
kd.contrib	Contrib public API.
kd.data	Data modules.
kd.data.iters	Dataset iterators.
kd.data.py	PyGrain public API.
kd.evals	Evaluator.
kd.from_xid	Helper for loading configs,… from XManager experiments.
kd.inspect	Inspect utils.
kd.kdash	Small library for creating Flatboard dashboards.
kd.knn	Wrapper aound flax.linen.Module to add torch-like API.
kd.konfig	Wrapper around ConfigDict to support auto-complete/type checking.
kd.konfig._default_values	Default values and configuration.
kd.kontext	Kontext is a small self-contained library to manipulate nested trees.
kd.losses	Losses.
kd.metrics	Metrics.
kd.nn	Collection of nn.Modules to build neural networks.
kd.optim	Optimizers etc.
kd.random	Small wrapper around jax.random for OO API.
kd.summaries	Summaries.
kd.summaries.deprecated	Deprecated summaries.
kd.testing	Testing utilities.
kd.train	Train.
kd.typing	Common Typing Annotations.
kd.utils	Utils public API.
kd.xm	XManager utils.

Class

kd.ckpts.AbstractPartialLoader	Abstract class for partial checkpoint loaders.
kd.ckpts.Checkpointer	Wrapper around Orbax CheckpointManager.
kd.ckpts.MultiTransform	Transform which applies multiple transformations sequentially.
kd.ckpts.NoopCheckpointer	Does nothing.
kd.ckpts.NoopTransform	init_transform that does nothing.
kd.ckpts.PartialKauldronLoader	Partial loader for Kauldron checkpoints.
kd.ckpts.items.CheckpointItem	Interface for a checkpoint item.
kd.ckpts.items.StandardCheckpointItem	Standard checkpoint item (for arbitrary jax.Array pytree).
kd.ckpts.items.TopLevelCheckpointItem	Checkpoint item that contains other sub-checkpoint items.
kd.data.AddConstants	Adds constant elements.
kd.data.BatchSize	Batch size.
kd.data.Cast	Cast an element to the specified dtype.
kd.data.ElementWiseTransform	Base class for elementwise transforms.
kd.data.Elements	Modify the elements by keeping xor dropping and/or renaming and/or copying.
kd.data.FilterTransform	Abstract base class for filter transformations for individual elements.
kd.data.Gather	Gathers entries along a single dimension.
kd.data.InMemoryPipeline	Pipeline which fit in memory.
kd.data.IterableDataset	General interface for iterable datasets.
kd.data.MapTransform	Abstract base class for all 1:1 transformations of elements.
kd.data.Pipeline	Base class for kauldron data pipelines.
kd.data.Rearrange	Einops rearrange on a single element.
kd.data.Resize	Resizes an image.
kd.data.TreeFlattenWithPath	Flatten any tree-structured elements.
kd.data.ValueRange	Map the value range of an element.
kd.data.iters.Iterator	Wrapper around a dataset iterator.
kd.data.iters.NonCheckpointableIterator	Handler that is not-checkpointable.
kd.data.iters.PyGrainIterator	PyGrain iterator.
kd.data.iters.TFDataIterator	Checkpointable tf.data iterator.
kd.data.py.AddConstants	Adds constant elements.
kd.data.py.Cast	Cast an element to the specified dtype.
kd.data.py.DataSource	Generic loader of arbitrary grain data source.
kd.data.py.DataSourceBase	Base class to implement a data source.
kd.data.py.ElementWiseTransform	Base class for elementwise transforms.
kd.data.py.Elements	Modify the elements by keeping xor dropping and/or renaming and/or copying.
kd.data.py.Gather	Gathers entries along a single dimension.
kd.data.py.HuggingFace	HuggingFace loader.
kd.data.py.Json	Json pipeline.
kd.data.py.Mix	Create a dataset mixture.
kd.data.py.PyGrainPipeline	Abstract base class to construct PyGrain data pipeline.
kd.data.py.Rearrange	Einops rearrange on a single element.
kd.data.py.Resize	Resizes an image.
kd.data.py.SliceDataset	Transform which select a subset of the dataset.
kd.data.py.Tfds	Base TFDS loader.
kd.data.py.TreeFlattenWithPath	Flatten any tree-structured elements.
kd.data.py.ValueRange	Map the value range of an element.
kd.evals.CollectionKeys	Names of the metrics/summaries/losses (displayed in flatboard).
kd.evals.Evaluator	Evaluator running num_batches times.
kd.evals.EvaluatorBase	Base class for inline evaluators.
kd.evals.EveryNSteps	Run eval every N train steps.
kd.evals.FewShotEvaluator	FewShotEvaluator running closed-form few-shot classification.
kd.evals.Once	Run eval only after the XX train steps.
kd.evals.RunStrategy	Base class for info on how to run the evaluation.
kd.evals.StandaloneEveryCheckpoint	Run eval continuously everytime a new checkpoint is found.
kd.evals.StandaloneLastCheckpoint	Run eval only after the last checkpoint, after train has completed.
kd.inspect.Profiler	kd.inspect.Profiler.
kd.kdash.BuildContext	Context for building the dashboard.
kd.kdash.DashboardsBase	Flatboard dashboard structure.
kd.kdash.MetricDashboards	Standard metrics & losses dashboards for a single collection.
kd.kdash.MultiDashboards	Container of multiple dashboards.
kd.kdash.NoopDashboard	Empty dashboard.
kd.kdash.Plot	Single plot inside a dashboard.
kd.kdash.SingleDashboard	Single dashboard containing multiple plots.
kd.knn.Dense	Dense(features: int, use_bias: bool = True, dtype: Union[str, type[Any], numpy.dtype, jax._src.typing.SupportsDType, Any, NoneType] = None, param_dtype: Union[str, type[Any], numpy.dtype, jax._src.typing.SupportsDType, Any] = <class ‘jax.numpy.float32’>, precision: Union[NoneType, str, jax._src.lax.lax.Precision, tuple[str, str], tuple[jax._src.lax.lax.Precision, jax._src.lax.lax.Precision]] = None, kernel_init: Union[jax.nn.initializers.Initializer, collections.abc.Callable[…, Any]] = <function variance_scaling..init at 0x7824acfd7f60>, bias_init: Union[jax.nn.initializers.Initializer, collections.abc.Callable[…, Any]] = <function zeros at 0x7824b06c2f20>, promote_dtype: flax.linen.linear.PromoteDtypeFn = <function promote_dtype at 0x7824ad016d40>, dot_general: collections.abc.Callable[…, typing.Union[jax.Array, typing.Any]]
kd.knn.Dropout	Dropout(rate: float, broadcast_dims: collections.abc.Sequence[int] = (), deterministic: bool
kd.knn.Module	Base Module class.
kd.knn.Sequential	Sequential(layers: collections.abc.Sequence[collections.abc.Callable[…, typing.Any]], parent: Union[flax.linen.module.Module, flax.core.scope.Scope, flax.linen.module._Sentinel, NoneType] = <flax.linen.module._Sentinel object at 0x7824acff2d90>, name: Optional[str] = None, *, _kd_state: ‘Optional[_ModuleState]’ = None)
kd.konfig.ConfigDict	Wrapper around ConfigDict.
kd.konfig.WithRef	Protocol to better access lazy fields.
kd.kontext.Context
kd.kontext.GlobPath	Represents a string path.
kd.kontext.Path	Represents a (non-glob) string path.
kd.losses.AbsoluteValue	Absolute value loss.
kd.losses.Huber	Huber loss.
kd.losses.L1	L1 loss.
kd.losses.L2	L2 loss.
kd.losses.Loss	Base class for losses which handles masks, averaging, and loss-weight.
kd.losses.NegativeCosineSimilarity	Negative Cosine Similarity loss.
kd.losses.SigmoidBinaryCrossEntropy	Sigmoid cross-entropy loss with binary labels.
kd.losses.SoftmaxCrossEntropy	Softmax cross-entropy loss.
kd.losses.SoftmaxCrossEntropyWithIntLabels	Softmax cross-entropy loss with integer labels.
kd.losses.Value	Value loss.
kd.metrics.Accuracy	Classification Accuracy.
kd.metrics.Ari	Adjusted Rand Index (ARI) computed from predictions and labels.
kd.metrics.AutoState	Flexible base class for conveniently defining custom states.
kd.metrics.AverageState	Computes the average of a scalar or a batch of tensors.
kd.metrics.BinaryAccuracy	Classification Accuracy for Binary classification tasks.
kd.metrics.CollectFirstState	Get the first outputs (possibly) across multiple steps (no reducing).
kd.metrics.CollectingState	Accumulate outputs across multiple steps (without reducing).
kd.metrics.EmptyState	Empty state.
kd.metrics.LpipsVgg	VGG LPIPS.
kd.metrics.Metric	Base class for metrics.
kd.metrics.NoopMetric	Metric that does nothing. Can be used in sweeps to remove a metric.
kd.metrics.Norm	Wraps jnp.linalg.norm to compute the average norm for given tensors.
kd.metrics.Precision1	Precision@1 for multilabel classification.
kd.metrics.Psnr	PSNR.
kd.metrics.RocAuc	Area Under the Receiver Operating Characteristic Curve (ROC AUC).
kd.metrics.SingleDimension	Returns a single chosen dimension of the tensor.
kd.metrics.SkipIfMissing	Skip this metric if any of the keys are missing.
kd.metrics.Ssim	Structural similarity (SSIM).
kd.metrics.State	Base metric state class.
kd.metrics.Std	Compute the standard deviation for float values.
kd.metrics.TreeMap	Maps an inner metric to a pytree and returns a pytree of results.
kd.metrics.TreeReduce	Applies a metric to a pytree and returns the aggregated result.
kd.nn.AddEmbedding	Helper Module for adding a PositionEmbedding e.g. in a knn.Sequential.
kd.nn.AddLearnedEmbedding	Adds learned positional embeddings to the inputs.
kd.nn.AttentionModule	Interface specification for Attention modules.
kd.nn.Dropout	Wrapper around nn.Dropout but using kd.nn.train_property.
kd.nn.DummyModel	Empty model that ignores inputs and always produces a single logit of 42.
kd.nn.ExternalModule	Module that is defined outside Kauldron.
kd.nn.FlatAutoencoder	Very simple auto-encoder class to showcase using keys and submodules.
kd.nn.FourierEmbedding	Apply Fourier position embedding to a grid of coordinates.
kd.nn.Identity	Module that applies the identity function to a single tensor.
kd.nn.ImageTokenizer	Interface for modules that convert images into tokens.
kd.nn.ImprovedMultiHeadDotProductAttention	Multi-head dot-product attention.
kd.nn.LearnedEmbedding	Learned positional embeddings.
kd.nn.MultiHeadDotProductAttention	Wrapper around nn.MultiHeadDotProductAttention using knn.train_property.
kd.nn.NormModule	Interface specification for norm modules (to be used as type annotation).
kd.nn.ParallelAttentionBlock	Parallel self attention (see Vit22B paper: arxiv.org/abs/2302.05442).
kd.nn.Patchify	Patchify an image, as in ViT (without linear embedding).
kd.nn.PatchifyEmbed	Patchify and linearly embed and image, as in ViT.
kd.nn.PostNormBlock	Post-LN Transformer layer (not recommended).
kd.nn.PreNormBlock	Pre-LN Transformer layer (default transformer layer).
kd.nn.Rearrange	Wrapper around einops.rearrange for usage e.g. in nn.Sequential.
kd.nn.Reduce	Wrapper around einops.reduce for usage e.g. in nn.Sequential.
kd.nn.Sequential	Like nn.Sequential but allows configuring input and output keys.
kd.nn.TransformerBlock	Interface definition for transformer blocks (for use in type annotations).
kd.nn.TransformerMLP	Simple MLP with a single hidden layer for use in Transformer blocks.
kd.nn.Vit	Basic Vision Transformer classifer with GAP.
kd.nn.VitEncoder	Basic Vit Encoder.
kd.nn.WrapperModule	Base class to wrapper a module.
kd.nn.ZeroEmbedding	Embedding that returns zero (for deactivating position embeddings).
kd.random.PRNGKey	Small wrapper around jax.random key arrays to reduce boilerplate.
kd.summaries.Histogram	Output type for histogram summaries.
kd.summaries.HistogramSummary	Basic histogram summary.
kd.summaries.PointCloud	Output type for point cloud summaries.
kd.summaries.ShowBoxes	Show a set of boxes with optional image reshaping.
kd.summaries.ShowDifferenceImages	Show a set of difference images with optional reshaping.
kd.summaries.ShowImages	Show image summaries with optional reshaping.
kd.summaries.ShowPointCloud	Show a point cloud with optional reshaping.
kd.summaries.ShowSegmentations	Show a set of segmentations with optional reshaping.
kd.summaries.ShowTexts	Show texts.
kd.summaries.deprecated.ImageSummary	Deprecated ImageSummary. Raises an error if instantiated.
kd.summaries.deprecated.Summary	Deprecated Summaries. Raises an error if instantiated.
kd.train.Auxiliaries	Wrapper around the losses, summaries and metrics.
kd.train.AuxiliariesOutput	Auxiliaries final values (after merge and compute).
kd.train.AuxiliariesState	Auxiliaries (intermediate states to be accumulated).
kd.train.Context	Namespace for retrieving information with path-based keys.
kd.train.RngStream	Info on one rng stream.
kd.train.RngStreams	Manager of rng streams.
kd.train.Setup	Setup/environment options.
kd.train.TqdmInfo	TqdmInfo(*, desc: ‘str’ = ‘train’, log_xm: ‘bool’ = True)
kd.train.TrainState	Data structure for checkpointing the model.
kd.train.TrainStep	Base Training Step.
kd.train.Trainer	Base trainer class.
kd.typing.Any
kd.typing.Array
kd.typing.ArraySpec	Describes an array via it’s dtype and shape.
kd.typing.AxisName
kd.typing.Bool
kd.typing.Complex
kd.typing.Complex64
kd.typing.Float
kd.typing.Float32
kd.typing.Hashable
kd.typing.Int
kd.typing.Integer
kd.typing.Memo	Jaxtyping information about the shapes in the current scope.
kd.typing.Num
kd.typing.Sequence	All the operations on a read-only sequence.
kd.typing.Shape	Helper to construct concrete shape tuples from shape-specs.
kd.typing.TfArray
kd.typing.TfFloat
kd.typing.TfFloat32
kd.typing.TfInt
kd.typing.TfUInt8
kd.typing.TypeCheckError	Indicates a runtime typechecking error from the @typechecked decorator.
kd.typing.UInt32
kd.typing.UInt8
kd.typing.XArray
kd.xm.Experiment	XManager experiment wrapper.
kd.xm.WorkUnit	XManager work unit wrapper.

Function

kd._filter_logs.add_filter	Add a filter to the absl logging handler.
kd.ckpts.workdir_from_xid
kd.from_xid.get_cfg	Returns the config/sub-config from an xmanager experiment.
kd.from_xid.get_element_spec	Returns the element_spec of the train dataset of an xmanager experiment.
kd.from_xid.get_resolved	Returns the resolved config/sub-config from an xmanager experiment.
kd.from_xid.get_workdir	Returns the workdir of an xmanager experiment.
kd.inspect.get_batch_stats	Return pd.DataFrame containing the batch stats.
kd.inspect.get_colab_model_overview	Return pd.DataFrame for displaying the model params, inputs,…
kd.inspect.get_connection_graph	Build the graphviz.
kd.inspect.json_spec_like	Convert etree.spec_like output to json and displays it in colab form.
kd.inspect.lower_trainstep	Returns lowered trainerstep.step.
kd.inspect.plot_batch	Display batch images.
kd.inspect.plot_context	Display the context structure.
kd.inspect.plot_schedules	Overview plot for (nested) dict of schedules.
kd.inspect.plot_sharding	Plot sharding.
kd.inspect.show_trainer_info	Display various plot on the trainer.
kd.kdash.build_and_upload	Create the dashboards.
kd.knn.convert	Decorator that convert a flax class into klinen.
kd.konfig.DEFINE_config_file	Defines flag for ConfigDict.
kd.konfig.imports	Contextmanager which replace import statements by configdicts.
kd.konfig.mock_modules	Contextmanager which replaces list of modules with ConfigDictProxyObjects.
kd.konfig.placeholder	Defines an entry in a ConfigDict that has no value yet.
kd.konfig.ref_copy	One-way recursive copy of the ConfigDict.
kd.konfig.ref_fn	Wrap a function for lazy-evaluation.
kd.konfig.register_aliases	Register module aliases for nicer display.
kd.konfig.register_default_values	Register default values when creating the ConfigDict.
kd.konfig.required	Defines a required attribute in the config that has no value yet.
kd.konfig.resolve	Recursively parses a nested ConfigDict and resolves module constructors.
kd.konfig.set_lazy_imported_modules	Set which modules inside with konfig.imports() will be lazy-imported.
kd.kontext.filter_by_path	Filters a context by a path.
kd.kontext.flatten_with_path	Flatten any PyTree / ConfigDict into a dict with ‘keys.like[0].this’.
kd.kontext.get_by_path	Get (nested) item or attribute by given path.
kd.kontext.get_keypaths	Return a dictionary mapping Key-annotated fieldnames to their paths.
kd.kontext.is_key_annotated	Check if a given class or instance has fields annotated with Key.
kd.kontext.path_builder_from	Create a path builder from a class.
kd.kontext.resolve_from_keyed_obj	Resolve the Key annotations of an object for given context.
kd.kontext.resolve_from_keypaths	Get values for key_paths from context with useful errors when failing.
kd.kontext.set_by_path	Mutate the obj to set the value.
kd.losses.compute_losses	Compute all losses based on given context.
kd.metrics.concat_field	Defines a AutoState data-field that is merged by concatenation.
kd.metrics.static_field	Define an AutoState static field.
kd.metrics.sum_field	Define an AutoState data-field that is merged by summation (a + b).
kd.metrics.truncate_field	Defines a AutoState data-field that is merged by truncation.
kd.nn.convert_to_fourier_features	Convert inputs to Fourier features, e.g. for positional encoding.
kd.nn.interms_property	interms property that makes storing intermediates more convenient.
kd.nn.set_train_property	Set the self.is_training state to the given value.
kd.nn.train_property	is_training property.
kd.optim.decay_to_init	Add (params - init_params) scaled by weight_decay.
kd.optim.exclude	Create a mask which selects all nodes except the ones matching the pattern.
kd.optim.named_chain	Wraps optax.named_chain and allows passing transformations as kwargs.
kd.optim.partial_updates	Applies the optimizer to a subset of the parameters.
kd.optim.select	Create a mask which selects only the sub-pytree matching the pattern.
kd.testing.assert_step_specs	Check the train step run correctly (fast).
kd.train.forward	Forward pass of the model.
kd.train.forward_with_loss	Forward pass of the model, including losses.
kd.typing.Dim	Helper to construct concrete Dim (for single-axis Shape).
kd.typing.check_type	Ensure that value matches expected_type, alias for typeguard.check_type.
kd.typing.enable_kd_type_checking	Enable custom type checking for Kauldron types.
kd.typing.set_shape	Validates the given shape and sets any previously unknown shapes.
kd.typing.typechecked	Decorator to enable runtime type-checking and shape-checking.
kd.xm.add_colab_artifacts	Add a link to the kd-infer colab.
kd.xm.add_log_artifacts	Add XManager artifacts for easy access to the Python logs.
kd.xm.add_tags_to_xm	Add tags to the xmanager experiment.
kd.xm.load_config_from_path	Loads a config from a path.

Attribute

kd.kontext.REQUIRED
kd.sharding
kd.typing.Union
kd.xm.xmanager_api

Typing

kd.knn.Intermediate
kd.konfig.ConfigDictLike
kd.kontext.Key
kd.kontext.KeyTree
kd.typing.Axes
kd.typing.Callable
kd.typing.DType
kd.typing.ElementSpec
kd.typing.Initializer
kd.typing.PRNGKey
kd.typing.PRNGKeyLike
kd.typing.PyTree
kd.typing.Scalar
kd.typing.ScalarFloat
kd.typing.ScalarInt
kd.typing.Schedule