Model registry API - Gitlabhq

Tier: Free, Premium, Ultimate
Offering: GitLab.com, GitLab Self-Managed, GitLab Dedicated

Use this API to interact with the machine learning model registry.

The :model_version_id attribute in each endpoint accepts either a model version ID or a candidate run ID. For more information, see Model version and candidate IDs.

Download a machine learning model package file

Downloads a specified file from a machine learning model package.

plaintext

GET /api/v4/projects/:id/packages/ml_models/:model_version_id/files/(*path/):file_name

Supported attributes:

Attribute	Type	Required	Description
`id`	integer or string	Yes	The ID or URL-encoded path of the project.
`model_version_id`	integer or string	Yes	The model version ID or candidate run ID. See Model version and candidate IDs.
`file_name`	string	Yes	The filename.
`path`	string	No	The directory path for the file.

If successful, returns 200 OK and the file contents.

Example request:

shell

curl --header "Authorization: Bearer <your_access_token>" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/2/files/foo.txt"

Example request with a directory path:

shell

curl --header "Authorization: Bearer <your_access_token>" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/2/files/my_dir/foo.txt"

Upload a model package file

Uploads a file to a machine learning model package.

Authorize the upload

Authorizes a file upload to a machine learning model package.

plaintext

PUT /api/v4/projects/:id/packages/ml_models/:model_version_id/files/(*path/):file_name/authorize

Supported attributes:

Attribute	Type	Required	Description
`id`	integer or string	Yes	The ID or URL-encoded path of the project.
`model_version_id`	integer or string	Yes	The model version ID or candidate run ID. See Model version and candidate IDs.
`file_name`	string	Yes	The filename.
`path`	string	No	The directory path for the file.

If successful, returns 200 OK.

Example request:

shell

curl --request PUT \
  --header "Authorization: Bearer <your_access_token>" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/2/files/model.pkl/authorize"

Send the file

Uploads the file to a machine learning model package.

plaintext

PUT /api/v4/projects/:id/packages/ml_models/:model_version_id/files/(*path/):file_name

Supported attributes:

Attribute	Type	Required	Description
`id`	integer or string	Yes	The ID or URL-encoded path of the project.
`model_version_id`	integer or string	Yes	The model version ID or candidate run ID. See Model version and candidate IDs.
`file_name`	string	Yes	The filename.
`path`	string	No	The directory path for the file.
`file`	file	Yes	The file to upload.

If successful, returns 201 Created.

Example request:

shell

curl --request PUT \
  --header "Authorization: Bearer <your_access_token>" \
  --form "[email protected]" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/2/files/model.pkl"

Example request with a directory path:

shell

curl --request PUT \
  --header "Authorization: Bearer <your_access_token>" \
  --form "[email protected]" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/2/files/my_dir/model.pkl"

Model version and candidate IDs

The :model_version_id attribute accepts either a model version ID or a candidate run ID.

To find the model version ID, check the URL of the model version page. For example, in https://gitlab.example.com/my-namespace/my-project/-/ml/models/1/versions/5, the model version ID is 5.

shell

curl --header "Authorization: Bearer <your_access_token>" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/5/files/model.pkl"

To use a candidate run ID, prepend the internal ID of the candidate with candidate:. For example, in https://gitlab.example.com/my-namespace/my-project/-/ml/candidates/5, the value for :model_version_id is candidate:5.

shell

curl --header "Authorization: Bearer <your_access_token>" \
  --url "https://gitlab.example.com/api/v4/projects/1/packages/ml_models/candidate:5/files/model.pkl"