Command: training

The flexai training command manages Training Jobs: from starting a new Training Job, retrieving its logs, and inspecting its execution, to downloading its output artifacts, among other tasks.

Available subcommands

flexai training checkpoints - Lists out the checkpoints that were generated by a Training Job.
flexai training debug-ssh - Establishes an SSH connection to a running Training Job.
flexai training delete - Deletes a Training Job.
flexai training fetch - Fetches artifacts from a Training Job.
flexai training inspect - Displays detailed information about a Training Job.
flexai training list - Lists all the Training Jobs.
flexai training logs - Displays the logs from a Training Job.
flexai training run - Starts a new Training Job.
flexai training stop - Stops a Training Job.

Training Lifecycle Statuses

A Training Job will go through a subset of the following statuses during its lifecycle:

Status	Description	Is Terminal
`pending`	The initial status of a Training Job. It means the Training Job was successfully received by FlexAI and is waiting to be processed	false
`scheduling`	FlexAI is looking for a suitable Node to handle the training given its hardware requirements	false
`rejected`	A Node that meets the requirements specified for the Training Job could not be found or is unavailable at the moment	true
`building`	A Node suitable for the Training Job was found. The building process has started. FlexAI is gathering all the components required for the training, in particular, cloning the revision specified for the source repository and installing the required dependencies	false
`in progress`	The building process completed successfully. The required FlexAI compute resources are being allocated	false
`enqueued`	The required compute resources specified for the Training Job were not available at the time. The Training Job is temporarily put on hold and will be scheduled to start once the required resources are freed up	false
`succeeded`	The Training Job completed successfully. The entry point training script terminated with exit code 0. Output artifacts can be downloaded using the `training fetch` command	true
`failed`	A Training Job can fail because one of the following reasons: The `--repository-revision` could not be found. The `requirements.txt` file could not be found in the root directory of the repository. The Training Job failed. The Entry point training script terminated with an exit code above 0. The Training Job’s duration exceeded the time limit (`BackoffLimitExceeded`).	true
`stop in progress`	The Training Job stopping process initiation was requested by the user and it is being performed	false
`stopped`	The Training Job was successfully stopped. If the Training Job was stopped while is status was `scheduling`, `building`, or `enqueued`, then no GPU resources were allocated If it was stopped after the training status changed to `building`, then the hardware resources were allocated and eventually released after the Training Job was successfully stopped	true
`stop failed`	The process of stopping a the Training Job failed	true

A Training or Fine-tuning Job in a “terminal” status can be deleted using the flexai training delete command.

Getting Started

Inference

Fine-tuning

Training

Platform Services

Interactive Development

CLI

Console

Best Practices

FAQ

Blueprints

Available subcommands

Training Lifecycle Statuses

​Available subcommands

​Training Lifecycle Statuses

Available subcommands

Training Lifecycle Statuses