AllocationID is the ID of an allocation of a task. It is usually of the form TaskID.allocation_number, maybe with some other metadata if different typ

ResourcePoolName holds the name of the resource pool, and describes the input/output of several ResourceManager methods.

TaskID is the unique ID of a task among all tasks.

AccessScopeID is an identifier for an access scope.

Duration is a JSON (un)marshallable version of time.Duration.

UserID is the type for user IDs.

JobID is the unique ID of a job among all jobs.

All the From... methods expose the more abstract representation returned by resource managers (since they can't all necessary know resources at the gr

ID is an identifier for an agent.

Y is Defaultable.

MetricGroup denotes what custom group the metric is.

ID is a unique ID assigned to the containers of tasks when started in the cluster.

Map is a generic, thread-safe map to supersede usages of sync.Map.

Support the tf2.3+ feature of passing numpy or tf logs to each callback.

A tiny, hacky http(s) server for testing ship logs. It's about the same amount of code as subclassing the stdlib's SimpleHTTPRequestHandler,

ID the type of Device.ID.

ExtendedFloat64 handles serializing floats to JSON, including special cases for infinite values.

Wrong number of outputs is not Mergable.

Z is Mergable.

Remote HTTP-based call

contextKeys are used to identify the type of value in the context. Since these are string, it is possible to get a short description of the context ke

* Represents the List in the ColumnPickerMenu component

DataLoader is meant to contain a user's ``data.Dataset``, configuration for sampling data in batches, and performance configuration like

Wrong number of inputs is not Mergable.

Wrong type of inputs is not Mergable.

ShouldExit breaks out of the top-level workload sequencer loop from inside function calls.

Enum values for warnings when launching commands. - LAUNCH_WARNING_UNSPECIFIED: Default value - LAUNCH_WARNING_CURRENT_SLOTS_EXCEEDED: For a d

DoneMsg is what each thread of exectuion puts on the doneq when it finishes.

LogLevel specifies the level for a log. - LOG_LEVEL_UNSPECIFIED: Unspecified log level. - LOG_LEVEL_TRACE: A log level of TRACE. - LOG_LEV

Permissions is a list of permissions.

TokenID is the type for token IDs.

UnsupportedError is returned when an unsupported feature of a resource manager is used.

Users is a slice of User objects—primarily useful for its methods.

The current state of the task. - STATE_UNSPECIFIED: The task state is unknown. - STATE_PULLING: The task's base image is being pulled from the

ExitCode is the process exit code of the container.

List of permissions types. Value of the enum has 9xxxx for global only permissions. Permissions on the same object share the thousands place value lik

RequestID links all operations with the same ID to a single trial create request.

SessionID is the type for user session IDs.

Offer a similar interface as WorkloadResponseInterceptor, except let send() yield a whole progression of RUN_STEP and COMPUTE_VALIDATION_METR

A result in a Login scenario indicating a particular token should be returned.

Defines a Batch unit for specifying length to PyTorch trainers. Batch(int) values are treated as periods, e.g. Batch(100) will checkpoint/va

Dict[int "concurrency", Hist]

ExitCode is the process exit code of the container.

JSONObj is a JSON object that converts to a []byte in SQL queries.

KeyCache caches only primary keys. KeyCache is just enough caching to allow Stream to automatically reconnect.

This class contains the functions we expect in order to accurately carry out parallel training. For custom model parallel training, you need

Roles is a list of Role.

StorageBackendID is the ID for the storage backend. Storage backend ID isn't backfilled so checkpoints older than 0.27.1 won't have this. There are al

TokenStore is a class for reading/updating a persistent store of user authentication tokens. TokenStore can remember tokens for many users fo

Type is a string holding the type of the Device.

byteString marshals to a base64 encoded string for the content of the file.

The current state of the trial. see \dT+ trial_state in db - STATE_UNSPECIFIED: The trial is in an unspecified state. - STATE_ACTIVE: The tria

Subclass subprocess.CalledProcessError in order to have a __str__ method that includes the stderr of the det cli call that failed. That

A result in a Login scenario indicating a check of a particular token is expected.

ClusterName is the name of the cluster within which we want to send a request.

Collector is the thread that reads and parses lines from stdout or stderr. It will pass structured data to the logq, and will send a message

``core.Context`` is a simple composition of several component APIs, with the following public members: - ``.checkpoint``, a :class:`~Ch

``DetCallback`` integrates a training loop built around ``transformers.Trainer`` with the Determined cluster. It reports metrics, uploads ch

Determined gives access to Determined API objects. Arguments: master (string, optional): The URL of the Determined master. If

DispatchState the model 'DispatchState'

DistributedSampler will iterate through an underlying sampler and return samples which belong to this shard. DistributedSampler is diffe

``DownloadMode`` defines the calling behavior of the .download() and the .restore_path() methods of ``CheckpointContext``. When mode is