v0.9.0

MLFlowClient v0.9.0

Diff since v0.8.0

New features

• Gateway services now work on any MLflow 3.x server (removed Databricks-only restriction)
• updategatewaymodeldefinition and updategatewayendpoint now expose all REST API parameters
• WebhookEntity.BUDGET_POLICY enum value added
• Dict constructors for FileInfo and MultipartUploadCredential
• Webhook docs page added; missing entries added to existing doc pages

Bug fixes

• updaterun: run_name now defaults to missing; fixed Integer(missing) crash when status omitted
• detachmodelfromgatewayendpoint: changed from DELETE to POST
• testwebhook: fixed response key ("result" not "test_result")
• listwebhooks: empty page_token no longer sent (caused server INTERNAL_ERROR)
• searchexperiments, searchruns, searchregisteredmodels, searchmodelversions: same empty-string fix
• createexperiment: workaround for MLflow 3.x basic-auth permission race condition
• GatewayEndpoint: Added experiment_id and usage_tracking fields
• Removed dead mlfpatch_v3 function

Breaking changes

• WebhookStatus.INACTIVE renamed to DISABLED
• WebhookEntity.PROMPT_VERSION_ALIAS renamed to PROMPT_ALIAS
• attachmodeltogatewayendpoint returns GatewayEndpointModelMapping instead of GatewayEndpoint
• Gateway type fields renamed (creation_timestamp to created_at, and update_timestamp to last_updated_at)
• GatewayEndpoint.model_configs renamed to model_mappings
• Search functions: filter/page_token type changed from String to Union{String,Missing}
• updategatewayendpoint: config param replaced with explicit API fields

Note

Previous version (v0.8.0) was tested against a mlflow==3.10 installation. An error occurred while updating, so apologies for the inconvenience.

Closed issues:

• artifact_location default is not being set (#48)
• Allow upgrade of JSON.jl compat (#72)
• Problems running with MLflow 3.9 (#76)
• TagBot: Manual intervention needed for releases (#77)

v0.8.0

MLFlowClient v0.8.0

Diff since v0.7.1

New features

• MLFlow 3.0 API Endpoints
  • Scorers API
  • Gateway API (Databricks-specific)
    • Secrets management
    • Model definitions
    • Endpoints
    • Endpoint bindings
    • Budgets
  • Prompt optimization API
  • MLFlow Artifacts API

Modifications (breaking changes)

• logmetric
  • Added parameters: model_id, dataset_name, dataset_digest`

Notes

• Gateway API
  • Gateway endpoints are Databricks-specific features
  • Tests will automatically skip if not using a Databricks tracking URI
  • Warning message displayed when attempting to run gateway tests on non-Databricks servers
• Prompt optimization API
  • Requires evaluation datasets to be present on the server
  • Tests gracefully skip with warnings if required resources are not available
• MLFlow Artifacts API
  • Uses mlflow-artifacts prefix (separate from main MLFlow API)
  • Multipart upload support for large files

Merged pull requests:

• Fix tracking URI to explicitly append /api to map to REST route (#78) (@aryawadhwa)
• 👽 Handle missing run output in createrun response (#80) (@Alexsp32)

v0.7.1

MLFlowClient v0.7.1

Diff since v0.7.0

Merged pull requests:

• Fix createexperiment API syntax in tutorial and examples (#65) (@RomeoV)
• Fix updaterun API syntax in tutorial and examples (#66) (@RomeoV)
• Fix logmetric array handling in tutorial and examples (#67) (@RomeoV)
• Remove logartifact calls from tutorial and examples (#68) (@RomeoV)
• Extend [compat] JSON="0.21,1" (#74) (@ablaom)

Closed issues:

• Tutorial references missing logartifact function (#61)
• Tutorial uses incorrect createexperiment API syntax (#62)
• Tutorial attempts to log arrays with logmetric (type error) (#63)
• Tutorial uses incorrect updaterun API syntax (#64)

v0.7.0

MLFlowClient v0.7.0

Diff since v0.6.1

Breaking changes

• Allowing compatibility with mlflow v3.x.x
  • Upgraded types with last changes in the API specification
  • Implemented logmodel function

Merged pull requests:

• Pairing mlflow v3 (#60) (@pebeto)

v0.6.1

MLFlowClient v0.6.1

Diff since v0.6.0

• Two new parameters in MLFlow instance constructor: username and password.
  • These will be injected to the Authorization header for each call to the server.
  • In the case username, password, and the Authorization header are provided, the instantiation will throw an error indicating that you can't do that.
• Allowing MLFLOW_TRACKING_USERNAME and MLFLOW_TRACKING_PASSWORD environment variables to set the username and password respectively.

Merged pull requests:

• Auth using environment variables (#56) (@characat0)

Closed issues:

• Authenticating via REST API (#33)
• Integrate with DagsHub's MLflow API (#35)
• Document how to set tags when creating an experiment (#50)

v0.6.0

MLFlowClient v0.6.0

Diff since v0.5.1

• New project structure to ensure better maintainability.
• Adopting Blue Style to improve readability.
• Pair with all the types from MLflow REST API.
• Implement all the endpoints from MLflow latest release.
• Improved documentation with multiple dispatch definitions.
• Simplify function definitions.
• Simplify test definitions.

Important

This update comes with some breaking changes. Projects like MLJFlow.jl will need to update carefully. It will be the same with personal projects.

Merged pull requests:

• Change MLFlow instances URI to use trailing /api (#47) (@pebeto)
• Improve documentation clarity for tags variable and update type in function definition. (#52) (@vonpetersenn)
• Add settag (#53) (@dpaetzel)
• Pairing with mlflow release api definition (#59) (@pebeto)

Closed issues:

• Improve tag definition (#30)
• Attributes such as model and dataset (#32)
• Proposal to buffer service requests (#41)
• Can't create experiment with v0.5.1 (#46)
• Update to the 2.x API (#54)
• Throw more readable HTTP error. (#58)

v0.5.1

MLFlowClient v0.5.1

Diff since v0.4.7

• Moved baseuri to apiroot, allowing users to specify its own API route (improving compatibility with platforms)
• Removed healthcheck function due to lack of support. It is not working on every platform providing a mlflow instance
• Added a MLFLOW_TRACKING_URI check when running tests

Merged pull requests:

• Adding use_ajax flag to MLFlow instances (#44) (@pebeto)

Closed issues:

• Potential for collaboration? (#5)
• How to log model? (#24)
• Support GitLabs MLFlow backend (#43)

v0.4.7

MLFlowClient v0.4.7

Diff since v0.4.6

Addings

• URI default to ENV["MLFLOW_TRACKING_URI"] if exists, just as mlflow documentation mentions.

Merged pull requests:

• Added dev. container (#38) (@stemann)

Closed issues:

• main doesn't pass it's own test (#42)

v0.4.6

MLFlowClient v0.4.6

Diff since v0.4.5

(fix)

• Solving Documenter pipeline failing on building
• Increasing project maintainability
• Increasing code test coverage

Merged pull requests:

• Improving code base (#37) (@pebeto)

Closed issues:

• Issue to trigger new releases (#39)

v0.4.5

MLFlowClient v0.4.5

Diff since v0.4.4

Merged pull requests:

• Migrating from /api to /ajax-api endpoint (#36) (@pebeto)