Skip to main content

MLflow 2.5.0

· 2 min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

MLflow 2.5.0 includes several major features and improvements:

  • [MLflow AI Gateway] We are excited to announce the release of MLflow AI Gateway, a powerful tool designed to streamline the usage and management of various large language model (LLM) providers, such as OpenAI and Anthropic, within an organization. It offers a standardized interface that simplifies the interaction with these services and delivers centralized, secure management of credentials. To get started with MLflow AI Gateway, check out the docs at https://mlflow.org/docs/latest/gateway/index.html. (#8694, @harupy, @BenWilson2, @dbczumar)
  • [Auth] We are excited to announce the release of authentication and authorization support for MLflow Tracking and the MLflow Model Registry, providing integrated access control capabilities to both services. To get started, check out the docs at https://mlflow.org/docs/latest/auth/index.html. (#9000, #8975, #8626, #8837, #8841, @gabrielfu, @harupy)

Features:

Bug fixes:

  • [Security] Improve robustness to LFI attacks on Windows by enhancing path validation (#8999, @serena-ruan)
    • If you are using mlflow server or mlflow ui on Windows, we recommend upgrading to MLflow 2.5.0 as soon as possible.
  • [Scoring] Support nullable array type values as spark_udf return values (#9014, @WeichenXu123)
  • [Models] Revert cache deletion of system modules when adding custom model code to the system path (#8722, @trungn1)
  • [Models] add micro version to mlflow version pinning (#8687, @C-K-Loan)
  • [Artifacts] Prevent manually deleted artifacts from causing artifact garbage collection to fail (#8498, @PenHsuanWang)

Documentation updates:

  • [Docs] Update .push_model_to_sagemaker docs (#8851, @pdifranc)
  • [Docs] Fix invalid link for Azure ML documentation (#8800, @dunnkers)
  • [Artifacts / Docs / Models / Projects] Adds information on the OCI MLflow plugins for seamless integration with Oralce Cloud Infrastructure services. (#8707, @mrDzurb)

Deprecation:

  • [Models] Deprecate the gluon model flavor. The mlflow.gluon module will be removed in a future release. (#8968, @harupy)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.4.0

· 3 min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

MLflow 2.4.0 includes several major features and improvements

Features:

  • [Tracking] Introduce dataset tracking APIs: mlflow.data and mlflow.log_input() (#8186, @prithvikannan)
  • [Tracking] Add mlflow.log_table() and mlflow.load_table() APIs for logging evaluation tables (#8523, #8467, @sunishsheth2009)
  • [Tracking] Introduce mlflow.get_parent_run() fluent API (#8493, @annzhang-db)
  • [Tracking / Model Registry] Re-introduce faster artifact downloads on Databricks (#8352, @dbczumar; #8561, @harupy)
  • [UI] Add dataset tracking information to MLflow Tracking UI (#8602, @prithvikannan, @hubertzub-db)
  • [UI] Introduce Artifact View for comparing inputs, outputs, and metadata across models (#8602, @hubertzub-db)
  • [Models] Extend mlflow.evaluate() to support LLM tasks (#8484, @harupy)
  • [Models] Support logging subclasses of Chain and LLMChain in mlflow.langchain flavor (#8453, @liangz1)
  • [Models] Add support for LangChain Agents to the mlflow.langchain flavor (#8297, @sunishsheth2009)
  • [Models] Add a mlflow.sentence_transformers flavor for SentenceTransformers (#8479, @BenWilson2; #8547, @Loquats)
  • [Models] Add support for multi-GPU inference and efficient weight loading for mlflow.transformers flavor (#8448, @ankit-db)
  • [Models] Support the max_shard_size parameter in the mlflow.transformers flavor (#8567, @wenfeiy-db)
  • [Models] Add support for audio transcription pipelines in the mlflow.transformers flavor (#8464, @BenWilson2)
  • [Models] Add support for audio classification to mlflow.transformers flavor (#8492, @BenWilson2)
  • [Models] Add support for URI inputs in audio models logged with the mlflow.transformers flavor (#8495, @BenWilson2)
  • [Models] Add support for returning classifier scores in mlflow.transformers pyfunc outputs (#8512, @BenWilson2)
  • [Models] Support optional inputs in model signatures (#8438, @apurva-koti)
  • [Models] Introduce an mlflow.models.set_signature() API to set the signature of a logged model (#8476, @jerrylian-db)
  • [Models] Persist ONNX Runtime InferenceSession options when logging a model with mlflow.onnx.log_model() (#8433, @leqiao-1)

Bug fixes:

  • [Tracking] Terminate Spark callback server when Spark Autologging is disabled or Spark Session is shut down (#8508, @WeichenXu123)
  • [Tracking] Fix compatibility of mlflow server with Flask<2.0 (#8463, @kevingreer)
  • [Models] Convert mlflow.transformers pyfunc scalar string output to list of strings during batch inference (#8546, @BenWilson2)
  • [Models] Fix a bug causing outdated pyenv versions to be installed by mlflow models build-docker (#8488, @Hellzed)
  • [Model Registry] Remove aliases from storage when a Model Version is deleted (#8459, @arpitjasa-db)

Documentation updates:

  • [Docs] Publish a new MLOps Quickstart for model selection and deployment (#8462, @lobrien)
  • [Docs] Add MLflavors library to Community Model Flavors documentation (#8420, @benjaminbluhm)
  • [Docs] Add documentation for Registered Model Aliases (#8445, @arpitjasa-db)
  • [Docs] Fix errors in documented mlflow models CLI command examples (#8480, @vijethmoudgalya)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.3.2

· 2 min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

MLflow 2.3.2 is a patch release containing the following features, bug fixes and changes:

Features:

  • [Models] Add GPU support for transformers models pyfunc inference and serving (#8375, @ankit-db)
  • [Models] Disable autologging functionality for non-relevant models when training a transformers model (#8405, @BenWilson2)
  • [Models] Add support for preserving and overriding torch_dtype values in transformers pipelines (#8421, @BenWilson2)
  • [Models] Add support for Feature Extraction pipelines in the transformers flavor (#8423, @BenWilson2)
  • [Tracking] Add basic HTTP auth support for users, registered models, and experiments permissions (#8286, @gabrielfu)

Bug Fixes:

  • [Models] Fix inferred schema issue with Text2TextGeneration pipelines in the transformers flavor (#8391, @BenWilson2)
  • [Models] Change MLflow dependency pinning in logged models from a range value to an exact major and minor version (#8422, @harupy)

Documentation updates:

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.3.1

· One min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

MLflow 2.3.1 is a patch release containing bug fixes and a security patch for GHSA-83fm-w79m-64r5. If you are using mlflow server or mlflow ui, we recommend upgrading to MLflow 2.3.1 as soon as possible.

Security patches:

  • [Security] Fix critical LFI attack vulnerability by disabling the ability to provide relative paths in registered model sources (#8281, @BenWilson2)

Bug fixes:

  • [Tracking] Fix an issue causing file and model uploads to hang on Databricks (#8348, @harupy)
  • [Tracking / Model Registry] Fix an issue causing file and model downloads to hang on Databricks (#8350, @dbczumar)
  • [Scoring] Fix regression in schema enforcement for model serving when using the inputs format for inference (#8326, @BenWilson2)
  • [Model Registry] Fix regression in model naming parsing where special characters were not accepted in model names (#8322, @arpitjasa-db)
  • [Recipes] Fix card rendering with the pandas profiler to handle columns containing all null values (#8263, @sunishsheth2009)

MLflow 2.3.0

· 3 min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

We are happy to announce the availability of MLflow 2.3.0!

MLflow 2.3.0 includes several major features and improvements

Features:

Bug fixes:

  • [Models] Fix a bug with tensorflow autologging for models with multiple inputs (#8097, @jaume-ferrarons)
  • [Recipes] Fix a bug with Pandas 2.0 updates for profiler rendering of datetime types (#7925, @sunishsheth2009)
  • [Tracking] Prevent exceptions from being raised if a parameter is logged with an existing key whose value is identical to the logged parameter (#8038, @AdamStelmaszczyk)
  • [Tracking] Fix an issue with deleting experiments in the FileStore backend (#8178, @mariusschlegel)
  • [Tracking] Fix a UI bug where the "Source Run" field in the Model Version page points to an incorrect set of artifacts (#8156, @WeichenXu123)
  • [Tracking] Fix a bug wherein renaming a run reverts its current lifecycle status to UNFINISHED (#8154, @WeichenXu123)
  • [Tracking] Fix a bug where a file URI could be used as a model version source (#8126, @harupy)
  • [Projects] Fix an issue with MLflow projects that have submodules contained within a project (#8050, @kota-iizuka)
  • [Examples] Fix lightning hyperparameter tuning examples (#8039, @BenWilson2)
  • [Server-infra] Fix bug with Cache-Control headers for static server files (#8016, @jmahlik)

Documentation updates:

  • [Examples] Add a new and thorough example for the creation of custom model flavors (#7867, @benjaminbluhm)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.2.2

· One min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

We are happy to announce the availability of MLflow 2.2.2!

MLflow 2.2.2 is a patch release containing the following bug fixes:

  • [Model Registry] Allow source to be a local path within a run's artifact directory if a run_id is specified (#7993, @harupy)
  • [Model Registry] Fix a bug where a windows UNC path is considered a local path (#7988, @WeichenXu123)
  • [Model Registry] Disallow name to be a file path in FileStore.get_registered_model (#7965, @harupy)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.2.1

· One min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

We are happy to announce the availability of MLflow 2.2.1!

MLflow 2.2.1 is a patch release containing the following bug fixes:

  • [Model Registry] Fix a bug that caused too many results to be requested by default when calling MlflowClient.search_model_versions() (#7935, @dbczumar)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.2.0

· 4 min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

We are happy to announce the availability of MLflow 2.2.0!

MLflow 2.2.0 includes several major features and improvements

Features:

Bug fixes:

  • [Recipes] Fix dataset format validation in the ingest step for custom dataset sources (#7638, @sunishsheth2009)
  • [Recipes] Fix bug in identification of worst performing examples during training (#7658, @sunishsheth2009)
  • [Recipes] Ensure consistent rendering of the recipe graph when inspect() is called (#7852, @sunishsheth2009)
  • [Recipes] Correctly respect positive_class configuration in the transform step (#7626, @sunishsheth2009)
  • [Recipes] Make logged metric names consistent with mlflow.evaluate() (#7613, @sunishsheth2009)
  • [Recipes] Add run_id and artifact_path keys to logged MLmodel files (#7651, @sunishsheth2009)
  • [UI] Fix bugs in UI validation of experiment names, model names, and tag keys (#7818, @subramaniam02)
  • [Tracking] Resolve artifact locations to absolute paths when creating experiments (#7670, @bali0019)
  • [Tracking] Exclude Delta checkpoints from Spark datasource autologging (#7902, @harupy)
  • [Tracking] Consistently return an empty list from GetMetricHistory when a metric does not exist (#7589, @bali0019; #7659, @harupy)
  • [Artifacts] Fix support for artifact operations on Windows paths in UNC format (#7750, @bali0019)
  • [Artifacts] Fix bug in HDFS artifact listing (#7581, @pwnywiz)
  • [Model Registry] Disallow creation of model versions with local filesystem sources in mlflow server (#7908, @harupy)
  • [Model Registry] Fix handling of deleted model versions in FileStore (#7716, @harupy)
  • [Model Registry] Correctly initialize Model Registry SQL tables independently of MLflow Tracking (#7704, @harupy)
  • [Models] Correctly move PyTorch model outputs from GPUs to CPUs during inference with pyfunc (#7885, @ankit-db)
  • [Build] Fix compatiblility issues with Python installations compiled using PYTHONOPTIMIZE=2 (#7791, @dbczumar)
  • [Build] Fix compatibility issues with the upcoming pandas 2.0 release (#7899, @harupy; #7910, @dbczumar)

Documentation updates:

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.1.1

· One min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

We are happy to announce the availability of MLflow 2.1.1!

MLflow 2.1.1 is a patch release containing the following bug fixes:

  • [Scoring] Fix mlflow.pyfunc.spark_udf() type casting error on model with ColSpec input schema and make PyFuncModel.predict support dataframe with elements of numpy.ndarray type (#7592 @WeichenXu123)
  • [Scoring] Make mlflow.pyfunc.scoring_server.client.ScoringServerClient support input dataframe with elements of numpy.ndarray type (#7594 @WeichenXu123)
  • [Tracking] Ensure mlflow imports ML packages lazily (#7597, @harupy)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.

MLflow 2.1.0

· 3 min read
MLflow maintainers
MLflow maintainers
MLflow maintainers

We are happy to announce the availability of MLflow 2.1.0!

MLflow 2.1.0 includes several major features and improvements

Features:

  • [Recipes] Introduce support for multi-class classification (#7458, @mshtelma)
  • [Recipes] Extend the pyfunc representation of classification models to output scores in addition to labels (#7474, @sunishsheth2009)
  • [UI] Add user ID and lifecycle stage quick search links to the Runs page (#7462, @jaeday)
  • [Tracking] Paginate the GetMetricHistory API (#7523, #7415, @BenWilson2)
  • [Tracking] Add Runs search aliases for Run name and start time that correspond to UI column names (#7492, @apurva-koti)
  • [Tracking] Add a /version endpoint to mlflow server for querying the server's MLflow version (#7273, @joncarter1)
  • [Model Registry] Add FileStore support for the Model Registry (#6605, @serena-ruan)
  • [Model Registry] Introduce an mlflow.search_registered_models() fluent API (#7428, @TSienki)
  • [Model Registry / Java] Add a getRegisteredModel() method to the Java client (#6602) (#7511, @drod331)
  • [Model Registry / R] Add an mlflow_set_model_version_tag() method to the R client (#7401, @leeweijie)
  • [Models] Introduce a metadata field to the MLmodel specification and log_model() methods (#7237, @jdonzallaz)
  • [Models] Extend Model.load() to support loading MLmodel specifications from remote locations (#7517, @dbczumar)
  • [Models] Pin the major version of MLflow in Models' requirements.txt and conda.yaml files (#7364, @BenWilson2)
  • [Scoring] Extend mlflow.pyfunc.spark_udf() to support StructType results (#7527, @WeichenXu123)
  • [Scoring] Extend TensorFlow and Keras Models to support multi-dimensional inputs with mlflow.pyfunc.spark_udf()(#7531, #7291, @WeichenXu123)
  • [Scoring] Support specifying deployment environment variables and tags when deploying models to SageMaker (#7433, @jhallard)

Bug fixes:

  • [Recipes] Fix a bug that prevented use of custom early_stop functions during model tuning (#7538, @sunishsheth2009)
  • [Recipes] Fix a bug in the logic used to create a Spark session during data ingestion (#7307, @WeichenXu123)
  • [Tracking] Make the metric names produced by mlflow.autolog() consistent with mlflow.evaluate() (#7418, @wenfeiy-db)
  • [Tracking] Fix an autologging bug that caused nested, redundant information to be logged for XGBoost and LightGBM models (#7404, @WeichenXu123)
  • [Tracking] Correctly classify SQLAlchemy OperationalErrors as retryable HTTP errors (#7240, @barrywhart)
  • [Artifacts] Correctly handle special characters in credentials when using FTP artifact storage (#7479, @HCTsai)
  • [Models] Address an issue that prevented MLeap models from being saved on Windows (#6966, @dbczumar)
  • [Scoring] Fix a permissions issue encountered when using NFS during model scoring with mlflow.pyfunc.spark_udf() (#7427, @WeichenXu123)

Documentation updates:

  • [Docs] Add more examples to the Runs search documentation page (#7487, @apurva-koti)
  • [Docs] Add documentation for Model flavors developed by the community (#7425, @mmerce)
  • [Docs] Add an example for logging and scoring ONNX Models (#7398, @Rusteam)
  • [Docs] Fix a typo in the model scoring REST API example for inputs with the dataframe_split format (#7540, @zhouyangyu)
  • [Docs] Fix a typo in the model scoring REST API example for inputs with the dataframe_records format (#7361, @dbczumar)

For a comprehensive list of changes, see the release change log, and check out the latest documentation on mlflow.org.