Experiment

Overview

Run vessl experiment --help to view the list of commands, vessl experiment [COMMAND] -help to view individual command instructions.

Create an experiment

1
vessl experiment create [OPTIONS]
Copied!
Option
Description
-c, --cluster
Cluster name (must be specified before other options)
-x, --command
Start command to execute in experiment container
-r, --resource
Resource type to run an experiment (for managed cluster only)
--processor
CPU or GPU (for custom cluster only)
--cpu-limit
Number of vCPUs (for custom cluster only)
--memory-limit
Memory limit in GiB (for custom cluster only)
--gpu-type
GPU type (for custom cluster only)
ex. Tesla-K80
--gpu-limit
Number of GPU cores (for custom cluster only)
--upload-local-file (multiple)
Upload local file. Format: [local_path] or [local_path]:[remote_path].
ex. --upload-local-file my-project:/home/vessl/my-project
--upload-local-git-diff
Upload local git commit hash and diff (only work in project repositories)
ex. --upload-local-git-diff=true
-i, --image-url
Kernel docker image URL
ex. vessl/kernels:py36.full-cpu
-m, --message
Message
--termination-protection
Enable termination protection
-e, --env-var (multiple)
Environment variables in the form of [key] [value]
ex. --env-var PORT 8080 --env-var DEBUG 1
--dataset (multiple)
Dataset mounts in the form of [mount_path] [dataset_name]
ex. --dataset /input mnist
--root-volume-size
Root volume size (defaults to 20Gi)
--working-dir
Working directory path (defaults to /work/[project_name])
--output-dir
Output directory path (defaults to /output
--local-project
Local project file URL
--worker-count
Number of workers (for distributed experiment only)
--framework-type
Specify pytorch or tensorflow(for distributed experiment only)

Download experiment output files

Each user can define experiment output files. You can save validation results, trained checkpoints, best performing models and other artifacts.
1
vessl experiment download-output [OPTIONS] NAME
Copied!
Argument
Description
NAME
Experiment name
Option
Description
-p, --path
Local download path (defaults to./output)
--worker-number
Worker number (for distributed experiment only)

List all experiments

1
vessl experiment list
Copied!

List experiment output files

Each user can define experiment output files. You can save validation results, trained checkpoints, best models, and other artifacts.
1
vessl experiment list-output [OPTIONS] NAME
Copied!
Argument
Description
NAME
Experiment name
Option
Description
-r, --recursive
List files recursively
--worker-number
Worker number (for distributed experiment only)
โ€‹

View logs of the experiment container

1
vessl experiment logs [OPTIONS] NAME
Copied!
Argument
Description
NAME
Experiment name
Option
Description
--tail
Number of lines to display from the end (defaults to 200)
--worker-number
Worker number (for distributed experiment only)
โ€‹

View information on the experiment

1
vessl experiment read NAME
Copied!
Argument
Description
NAME
Experiment name
Last modified 2mo ago