Environment Variables

AIchor makes a set of predefined environment variables available inside every experiment. Custom variables can also be defined in two ways: at the project level, applying to every experiment in a project, and at the experiment level, scoped to a single experiment through its manifest.

Project-level Environment variables

Project-level variables are defined from the AIchor UI and apply across all experiments in the project. From the AIchor project page, open the Variables tab and select Create new variable.

Two kinds of project-level variable can be created:

• Runtime variables — made available to the experiment's containers while it runs.
• Build-time variables — made available during the image build step.

A variable can also be marked as sensitive. The values of sensitive variables are not displayed in the UI after creation, which makes them suited to secrets such as API keys and access tokens. Build-time sensitive variables are the recommended way to pass secrets to an image build — see Docker Build Secrets.

Experiment-level Environment variables

Experiment-level variables are defined directly in the experiment manifest under spec.env. They are scoped to a single experiment, which makes them suited to values that change between runs, such as hyperparameters, without altering the shared project configuration.

Each entry is injected into the experiment as a plain key/value environment variable:

spec:
  operator: pytorch
  command: python train.py
  env:
    LEARNING_RATE: "0.001"
    BATCH_SIZE: "32"
    NUM_EPOCHS: "100"

The following rules apply:

• Values are injected literally, exactly as written in the manifest.
• When the same key is defined both under spec.env and as a project-level variable, the spec.env value takes precedence.
• The protected names listed below cannot be set through spec.env. Attempting to do so causes a validation error when the experiment is submitted.

Predefined Environment variables

The tables below list the environment variables that AIchor makes available inside every experiment, grouped by category, along with example values.

warning

Attempting to overwrite any protected variable causes an error during experiment execution.

Project and experiment

Environment Variable Name	Description	Example value	Overwritable
AICHOR_PROJECT_ID	AIchor project ID	e7be88af-7621-4574-aa61-e810679b69d4	❌
AICHOR_EXPERIMENT_ID	AIchor experiment ID	e2b53b98-975d-4be7-9f5b-8dfd6b765d32	❌
AICHOR_EXPERIMENT_MESSAGE	Message describing the experiment; defaults to the short commit message, or a custom message set at submission	Tune learning rate schedule	❌
AICHOR_ENGINE_NAME	Current engine name	AIchor	❌
AICHOR_CLUSTER_NAME	Current cluster name	AIchor	❌
AICHOR_CLUSTER_ID	Current cluster ID	31ed1de8-c937-40ea-b5c8-d17eecb12ec1	❌

Storage and buckets

Environment Variable Name	Description	Example value	Overwritable
AICHOR_INPUT_PATH	Input bucket path	s3://project-name-e7be88af76214574-inputs/	❌
AICHOR_INPUT_BUCKET_NAME	Input bucket name (without the s3:// prefix)	project-name-e7be88af76214574-inputs	❌
AICHOR_OUTPUT_PATH	Recommended output path (in the output bucket)	s3://project-name-e7be88af76214574-outputs/output/e2b53b98-975d-4be7-9f5b-8dfd6b765d32/	❌
AICHOR_OUTPUT_BUCKET_NAME	Output bucket name (without the s3:// prefix)	project-name-e7be88af76214574-outputs	❌
AICHOR_TENSORBOARD_PATH	Recommended path for TensorBoard records (in the output bucket); set only when TensorBoard is enabled in your manifest	s3://project-name-e7be88af76214574-outputs/output/e2b53b98-975d-4be7-9f5b-8dfd6b765d32/tensorboard/	❌
AWS_ENDPOINT_URL	S3 endpoint URL used by AWS SDK clients; usually points to the buckets AIchor created for the project	https://storage.googleapis.com	✅
AWS_ACCESS_KEY_ID	AWS key for configuring S3 automatically	AKIAIOSFODNN7EXAMPLE	✅
AWS_SECRET_ACCESS_KEY	AWS secret for configuring S3 automatically	wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY	✅
AWS_REQUEST_CHECKSUM_CALCULATION	Controls when a checksum is calculated for request payloads	when_required	❌
AWS_RESPONSE_CHECKSUM_VALIDATION	Controls when checksum validation is performed for response payloads	when_required	❌

Version control (Git)

A "slug" is a string that contains no special characters other than hyphens (-). Slugs are safe to use in URLs, Docker image names, and tags.

Environment Variable Name	Description	Example value	Overwritable
VCS_TYPE	VCS repository type	gitlab, github or bitbucket	❌
VCS_URL	URL of the remote Git repository	https://github.com/instadeep/group/subgroup/repo.git	❌
VCS_REPOSITORY	Full repository path, including owner, groups, and name	instadeep/group/subgroup/repo	❌
VCS_REPOSITORY_SLUG	Slugified VCS_REPOSITORY	instadeep-group-subgroup-repo	❌
VCS_REPOSITORY_OWNER	First segment of the repository URL, usually a username	instadeep	❌
VCS_REPOSITORY_OWNER_SLUG	Slugified VCS_REPOSITORY_OWNER	instadeep	❌
VCS_REPOSITORY_GROUPS	Intermediate segments of the repository URL	group/subgroup	❌
VCS_REPOSITORY_GROUPS_SLUG	Slugified VCS_REPOSITORY_GROUPS	group-subgroup	❌
VCS_REPOSITORY_NAME	Last segment of the repository URL	repo	❌
VCS_REPOSITORY_NAME_SLUG	Slugified VCS_REPOSITORY_NAME	repo	❌
VCS_REF	Git reference of the branch or tag	refs/heads/current-branch-name	❌
VCS_REF_SLUG	Slugified VCS_REF	refs-heads-current-branch-name	❌
VCS_REF_NAME	Name of the branch or tag	current-branch-name	❌
VCS_REF_NAME_SLUG	Slugified VCS_REF_NAME	current-branch-name	❌
VCS_SHA	Git commit SHA	ffac537e6cbbf934b08745a378932722df287a53	❌
VCS_SHA_SHORT	Shortened Git commit SHA	ffac537e	❌
VCS_AUTHOR_NAME	Git commit author name	Jane Doe	❌
VCS_AUTHOR_EMAIL	Git commit author email address	j.doe@instadeep.com	❌
VCS_COMMIT_MESSAGE	Full Git commit message	Tune learning rate schedule Lower the initial value for training stability	❌
VCS_COMMIT_MESSAGE_SHORT	First line of the Git commit message	Tune learning rate schedule	❌

User and organization

Environment Variable Name	Description	Example value	Overwritable
AICHOR_USER_NAME	Name of the user who launched the experiment	Jane Doe	❌
AICHOR_USER_EMAIL	Email address of the user who launched the experiment	j.doe@instadeep.com	❌
AICHOR_ORGANIZATION_NAME	Current organization name	instadeep	❌
AICHOR_ORGANIZATION_ID	Current organization ID	org_wa4GMsQSU1ZrZ921	❌

Compute and runtime

Environment Variable Name	Description	Example value	Overwritable
CURRENT_POD_IP	IP address of the pod in the cluster	10.110.20.26	❌
CPU_REQUEST	Number of CPUs requested	12	❌
NODE_NAME	Name of the node on which the pod is running	my-nvidia-gb200	❌