MLflow 2.10.0 includes several major features and improvements
In MLflow 2.10, we're introducing a number of significant new features that are preparing the way for current and future enhanced support for Deep Learning use cases, new features to support a broadened support for GenAI applications, and some quality of life improvements for the MLflow Deployments Server (formerly the AI Gateway).
Our biggest features this release are:
-
We have a new home. The new site landing page is fresh, modern, and contains more content than ever. We're adding new content and blogs all of the time.
-
Objects and Arrays are now available as configurable input and output schema elements. These new types are particularly useful for GenAI-focused flavors that can have complex input and output types. See the new Signature and Input Example documentation to learn more about how to use these new signature types.
-
LangChain has autologging support now! When you invoke a chain, with autologging enabled, we will automatically log most chain implementations, recording and storing your configured LLM application for you. See the new Langchain documentation to learn more about how to use this feature.
-
The MLflow
transformers
flavor now supports prompt templates. You can now specify an application-specific set of instructions to submit to your GenAI pipeline in order to simplify, streamline, and integrate sets of system prompts to be supplied with each input request. Check out the updated guide to transformers to learn more and see examples! -
The MLflow Deployments Server now supports two new requested features: (1) OpenAI endpoints that support streaming responses. You can now configure an endpoint to return realtime responses for Chat and Completions instead of waiting for the entire text contents to be completed. (2) Rate limits can now be set per endpoint in order to help control cost overrun when using SaaS models.
-
Continued the push for enhanced documentation, guides, tutorials, and examples by expanding on core MLflow functionality (Deployments, Signatures, and Model Dependency management), as well as entirely new pages for GenAI flavors. Check them out today!
Features:
- [Models] Introduce
Objects
andArrays
support for model signatures (#9936, @serena-ruan) - [Models] Support saving prompt templates for transformers (#10791, @daniellok-db)
- [Models] Enhance the MLflow Models
predict
API to serve as a pre-logging validator of environment compatibility. (#10759, @B-Step62) - [Models] Add support for Image Classification pipelines within the transformers flavor (#10538, @KonakanchiSwathi)
- [Models] Add support for retrieving and storing license files for transformers models (#10871, @BenWilson2)
- [Models] Add support for model serialization in the Visual NLP format for JohnSnowLabs flavor (#10603, @C-K-Loan)
- [Models] Automatically convert OpenAI input messages to LangChain chat messages for
pyfunc
predict (#10758, @dbczumar) - [Tracking] Add support for Langchain autologging (#10801, @serena-ruan)
- [Tracking] Enhance async logging functionality by ensuring flush is called on
Futures
objects (#10715, @chenmoneygithub) - [Tracking] Add support for a non-interactive mode for the
login()
API (#10623, @henxing) - [Scoring] Allow MLflow model serving to support direct
dict
inputs with themessages
key (#10742, @daniellok-db, @B-Step62) - [Deployments] Add streaming support to the MLflow Deployments Server for OpenAI streaming return compatible routes (#10765, @gabrielfu)
- [Deployments] Add the ability to set rate limits on configured endpoints within the MLflow deployments server API (#10779, @TomeHirata)
- [Deployments] Add support for directly interfacing with OpenAI via the MLflow Deployments server (#10473, @prithvikannan)
- [UI] Introduce a number of new features for the MLflow UI (#10864, @daniellok-db)
- [Server-infra] Add an environment variable that can disallow HTTP redirects (#10655, @daniellok-db)
- [Artifacts] Add support for Multipart Upload for Azure Blob Storage (#10531, @gabrielfu)
Bug fixes:
- [Models] Add deduplication logic for pip requirements and extras handling for MLflow models (#10778, @BenWilson2)
- [Models] Add support for paddle 2.6.0 release (#10757, @WeichenXu123)
- [Tracking] Fix an issue with an incorrect retry default timeout for urllib3 1.x (#10839, @BenWilson2)
- [Recipes] Fix an issue with MLflow Recipes card display format (#10893, @WeichenXu123)
- [Java] Fix an issue with metadata collection when using Streaming Sources on certain versions of Spark where Delta is the source (#10729, @daniellok-db)
- [Scoring] Fix an issue where SageMaker tags were not propagating correctly (#9310, @clarkh-ncino)
- [Windows / Databricks] Fix an issue with executing Databricks run commands from within a Window environment (#10811, @wolpl)
- [Models / Databricks] Disable
mlflowdbfs
mounts for JohnSnowLabs flavor due to flakiness (#9872, @C-K-Loan)
Documentation updates:
- [Docs] Fixed the
KeyError: 'loss'
bug for the Quickstart guideline (#10886, @yanmxa) - [Docs] Relocate and supplement Model Signature and Input Example docs (#10838, @BenWilson2)
- [Docs] Add the HuggingFace Model Evaluation Notebook to the website (#10789, @BenWilson2)
- [Docs] Rewrite the search run documentation (#10863, @chenmoneygithub)
- [Docs] Create documentation for transformers prompt templates (#10836, @daniellok-db)
- [Docs] Refactoring of the Getting Started page (#10798, @BenWilson2)
- [Docs] Add a guide for model dependency management (#10807, @B-Step62)
- [Docs] Add tutorials and guides for LangChain (#10770, @BenWilson2)
- [Docs] Refactor portions of the Deep Learning documentation landing page (#10736, @chenmoneygithub)
- [Docs] Refactor and overhaul the Deployment documentation and add new tutorials (#10726, @B-Step62)
- [Docs] Add a PyTorch landing page, quick start, and guide (#10687, #10737 @chenmoneygithub)
- [Docs] Add additional tutorials to OpenAI flavor docs (#10700, @BenWilson2)
- [Docs] Enhance the guides on quickly getting started with MLflow by demonstrating how to use Databricks Community Edition (#10663, @BenWilson2)
- [Docs] Create the OpenAI Flavor landing page and intro notebooks (#10622, @BenWilson2)
- [Docs] Refactor the Tensorflow flavor API docs (#10662, @chenmoneygithub)
Small bug fixes and documentation updates:
#10538, #10901, #10903, #10876, #10833, #10859, #10867, #10843, #10857, #10834, #10814, #10805, #10764, #10771, #10733, #10724, #10703, #10710, #10696, #10691, #10692, @B-Step62; #10882, #10854, #10395, #10725, #10695, #10712, #10707, #10667, #10665, #10654, #10638, #10628, @harupy; #10881, #10875, #10835, #10845, #10844, #10651, #10806, #10786, #10785, #10781, #10741, #10772, #10727, @serena-ruan; #10873, #10755, #10750, #10749, #10619, @WeichenXu123; #10877, @amueller; #10852, @QuentinAmbard; #10822, #10858, @gabrielfu; #10862, @jerrylian-db; #10840, @ernestwong-db; #10841, #10795, #10792, #10774, #10776, #10672, @BenWilson2; #10827, #10826, #10825, #10732, #10481, @michael-berk; #10828, #10680, #10629, @daniellok-db; #10799, #10800, #10578, #10782, #10783, #10723, #10464, @annzhang-db; #10803, #10731, #10708, @kriscon-db; #10797, @dbczumar; #10756, #10751, @Ankit8848; #10784, @AveshCSingh; #10769, #10763, #10717, @chenmoneygithub; #10698, @rmalani-db; #10767, @liangz1; #10682, @cdreetz; #10659, @prithvikannan; #10639, #10609, @TomeHirata
MLflow 2.9.2 is a patch release, containing several critical security fixes and configuration updates to support extremely large model artifacts.
Features:
- [Deployments] Add the
mlflow.deployments.openai
API to simplify direct access to OpenAI services through the deployments API (#10473, @prithvikannan) - [Server-infra] Add a new environment variable that permits disabling http redirects within the Tracking Server for enhanced security in publicly accessible tracking server deployments (#10673, @daniellok-db)
- [Artifacts] Add environment variable configurations for both Multi-part upload and Multi-part download that permits modifying the per-chunk size to support extremely large model artifacts (#10648, @harupy)
Security fixes:
- [Server-infra] Disable the ability to inject malicious code via manipulated YAML files by forcing YAML rendering to be performed in a secure Sandboxed mode (#10676, @BenWilson2, #10640, @harupy)
- [Artifacts] Prevent path traversal attacks when querying artifact URI locations by disallowing
..
path traversal queries (#10653, @B-Step62) - [Data] Prevent a mechanism for conducting a malicious file traversal attack on Windows when using tracking APIs that interface with
HTTPDatasetSource
(#10647, @BenWilson2) - [Artifacts] Prevent a potential path traversal attack vector via encoded url traversal paths by decoding paths prior to evaluation (#10650, @B-Step62)
- [Artifacts] Prevent the ability to conduct path traversal attacks by enforcing the use of sanitized paths with the tracking server (#10666, @harupy)
- [Artifacts] Prevent path traversal attacks when using an FTP server as a backend store by enforcing base path declarations prior to accessing user-supplied paths (#10657, @harupy)
Documentation updates:
- [Docs] Add an end-to-end tutorial for RAG creation and evaluation (#10661, @AbeOmor)
- [Docs] Add Tensorflow landing page (#10646, @chenmoneygithub)
- [Deployments / Tracking] Add endpoints to LLM evaluation docs (#10660, @prithvikannan)
- [Examples] Add retriever evaluation tutorial for LangChain and improve the Question Generation tutorial notebook (#10419, @liangz1)
Small bug fixes and documentation updates:
#10677, #10636, @serena-ruan; #10652, #10649, #10641, @harupy; #10643, #10632, @BenWilson2
MLflow 2.9.1 is a patch release, containing a critical bug fix related to loading pyfunc
models that were saved in previous versions of MLflow.
Bug fixes:
- [Models] Revert Changes to PythonModel that introduced loading issues for models saved in earlier versions of MLflow (#10626, @BenWilson2)
Small bug fixes and documentation updates:
#10625, @BenWilson2
MLflow 2.9.0 includes several major features and improvements.
MLflow AI Gateway deprecation (#10420, @harupy):
The feature previously known as MLflow AI Gateway has been moved to utilize the MLflow deployments API. For guidance on migrating from the AI Gateway to the new deployments API, please see the [MLflow AI Gateway Migration Guide](https://mlflow.org/docs/latest/llms/gateway/migration.html.
MLflow Tracking docs overhaul (#10471, @B-Step62):
The MLflow tracking docs have been overhauled. We'd like your feedback on the new tracking docs!
Security fixes:
Three security patches have been filed with this release and CVE's have been issued with the details involved in the security patch and potential attack vectors. Please review and update your tracking server deployments if your tracking server is not securely deployed and has open access to the internet.
- Sanitize
path
inHttpArtifactRepository.list_artifacts
(#10585, @harupy) - Sanitize
filename
inContent-Disposition
header forHTTPDatasetSource
(#10584, @harupy). - Validate
Content-Type
header to prevent POST XSS (#10526, @B-Step62)
Features:
- [Tracking] Use
backoff_jitter
when making HTTP requests (#10486, @ajinkyavbhandare) - [Tracking] Add default
aggregate_results
if the score type is numeric inmake_metric
API (#10490, @sunishsheth2009) - [Tracking] Add string type of score types for metric value for genai (#10307, @sunishsheth2009)
- [Artifacts] Support multipart upload for for proxy artifact access (#9521, @harupy)
- [Models] Support saving
torch_dtype
for transformers models (#10586, @serena-ruan) - [Models] Add built-in metric
ndcg_at_k
to retriever evaluation (#10284, @liangz1) - [Model Registry] Implement universal
copy_model_version
(#10308, @jerrylian-db) - [Models] Support saving/loading
RunnableSequence
,RunnableParallel
, andRunnableBranch
(#10521, #10611, @serena-ruan)
Bug fixes:
- [Tracking] Resume system metrics logging when resuming an existing run (#10312, @chenmoneygithub)
- [UI] Fix incorrect sorting order in line chart (#10553, @B-Step62)
- [UI] Remove extra whitespace in git URLs (#10506, @mrplants)
- [Models] Make spark_udf use NFS to broadcast model to spark executor on databricks runtime and spark connect mode (#10463, @WeichenXu123)
- [Models] Fix promptlab pyfunc models not working for chat routes (#10346, @daniellok-db)
Documentation updates:
- [Docs] Add a quickstart guide for Tensorflow (#10398, @chenmoneygithub)
- [Docs] Improve the parameter tuning guide (#10344, @chenmoneygithub)
- [Docs] Add a guide for system metrics logging (#10429, @chenmoneygithub)
- [Docs] Add instructions on how to configure credentials for Azure OpenAI (#10560, @BenWilson2)
- [Docs] Add docs and tutorials for Sentence Transformers flavor (#10476, @BenWilson2)
- [Docs] Add tutorials, examples, and guides for Transformers Flavor (#10360, @BenWilson2)
Small bug fixes and documentation updates:
#10567, #10559, #10348, #10342, #10264, #10265, @B-Step62; #10595, #10401, #10418, #10394, @chenmoneygithub; #10557, @dan-licht; #10584, #10462, #10445, #10434, #10432, #10412, #10411, #10408, #10407, #10403, #10361, #10340, #10339, #10310, #10276, #10268, #10260, #10224, #10214, @harupy; #10415, @jessechancy; #10579, #10555, @annzhang-db; #10540, @wllgrnt; #10556, @smurching; #10546, @mbenoit29; #10534, @gabrielfu; #10532, #10485, #10444, #10433, #10375, #10343, #10192, @serena-ruan; #10480, #10416, #10173, @jerrylian-db; #10527, #10448, #10443, #10442, #10441, #10440, #10439, #10381, @prithvikannan; #10509, @keenranger; #10508, #10494, @WeichenXu123; #10489, #10266, #10210, #10103, @TomeHirata; #10495, #10435, #10185, @daniellok-db; #10319, @michael-berk; #10417, @bbqiu; #10379, #10372, #10282, @BenWilson2; #10297, @KonakanchiSwathi; #10226, #10223, #10221, @milinddethe15; #10222, @flooxo; #10590, @letian-w;
MLflow 2.8.1 is a patch release, containing some critical bug fixes and an update to our continued work on reworking our docs.
Notable details:
- The API
mlflow.llm.log_predictions
is being marked as deprecated, as its functionality has been incorporated intomlflow.log_table
. This API will be removed in the 2.9.0 release. (#10414, @dbczumar)
Bug fixes:
- [Artifacts] Fix a regression in 2.8.0 where downloading a single file from a registered model would fail (#10362, @BenWilson2)
- [Evaluate] Fix the
Azure OpenAI
integration formlflow.evaluate
when using LLMjudge
metrics (#10291, @prithvikannan) - [Evaluate] Change
Examples
to optional for themake_genai_metric
API (#10353, @prithvikannan) - [Evaluate] Remove the
fastapi
dependency when usingmlflow.evaluate
for LLM results (#10354, @prithvikannan) - [Evaluate] Fix syntax issues and improve the formatting for generated prompt templates (#10402, @annzhang-db)
- [Gateway] Fix the Gateway configuration validator pre-check for OpenAI to perform instance type validation (#10379, @BenWilson2)
- [Tracking] Fix an intermittent issue with hanging threads when using asynchronous logging (#10374, @chenmoneygithub)
- [Tracking] Add a timeout for the
mlflow.login()
API to catch invalid hostname configuration input errors (#10239, @chenmoneygithub) - [Tracking] Add a
flush
operation at the conclusion of logging system metrics (#10320, @chenmoneygithub) - [Models] Correct the prompt template generation logic within the Prompt Engineering UI so that the prompts can be used in the Python API (#10341, @daniellok-db)
- [Models] Fix an issue in the
SHAP
model explainability functionality withinmlflow.shap.log_explanation
so that duplicate or conflicting dependencies are not registered when logging (#10305, @BenWilson2)
Documentation updates:
- [Docs] Add MLflow Tracking Quickstart (#10285, @BenWilson2)
- [Docs] Add tracking server configuration guide (#10241, @chenmoneygithub)
- [Docs] Refactor and improve the model deployment quickstart guide (#10322, @prithvikannan)
- [Docs] Add documentation for system metrics logging (#10261, @chenmoneygithub)
Small bug fixes and documentation updates:
#10367, #10359, #10358, #10340, #10310, #10276, #10277, #10247, #10260, #10220, #10263, #10259, #10219, @harupy; #10313, #10303, #10213, #10272, #10282, #10283, #10231, #10256, #10242, #10237, #10238, #10233, #10229, #10211, #10231, #10256, #10242, #10238, #10237, #10229, #10233, #10211, @BenWilson2; #10375, @serena-ruan; #10330, @Haxatron; #10342, #10249, #10249, @B-Step62; #10355, #10301, #10286, #10257, #10236, #10270, #10236, @prithvikannan; #10321, #10258, @jerrylian-db; #10245, @jessechancy; #10278, @daniellok-db; #10244, @gabrielfu; #10226, @milinddethe15; #10390, @bbqiu; #10232, @sunishsheth2009
MLflow 2.8.0 includes several notable new features and improvements
- The MLflow Evaluate API has had extensive feature development in this release to support LLM workflows and multiple new evaluation modalities. See the new documentation, guides, and tutorials for MLflow LLM Evaluate to learn more.
- The MLflow Docs modernization effort has started. You will see a very different look and feel to the docs when visiting them, along with a batch of new tutorials and guides. More changes will be coming soon to the docs!
- 4 new LLM providers have been added! Google PaLM 2, AWS Bedrock, AI21 Labs, and HuggingFace TGI can now be configured and used within the AI Gateway. Learn more in the new AI Gateway docs!
Features:
- [Gateway] Add support for AWS Bedrock as a provider in the AI Gateway (#9598, @andrew-christianson)
- [Gateway] Add support for Huggingface Text Generation Inference as a provider in the AI Gateway (#10072, @SDonkelaarGDD)
- [Gateway] Add support for Google PaLM 2 as a provider in the AI Gateway (#9797, @arpitjasa-db)
- [Gateway] Add support for AI21labs as a provider in the AI Gateway (#9828, #10168, @zhe-db)
- [Gateway] Introduce a simplified method for setting the configuration file location for the AI Gateway via environment variable (#9822, @danilopeixoto)
- [Evaluate] Introduce default provided LLM evaluation metrics for MLflow evaluate (#9913, @prithvikannan)
- [Evaluate] Add support for evaluating inference datasets in MLflow evaluate (#9830, @liangz1)
- [Evaluate] Add support for evaluating single argument functions in MLflow evaluate (#9718, @liangz1)
- [Evaluate] Add support for Retriever LLM model type evaluation within MLflow evaluate (#10079, @liangz1)
- [Models] Add configurable parameter for external model saving in the ONNX flavor to address a regression (#10152, @daniellok-db)
- [Models] Add support for saving inference parameters in a logged model's input example (#9655, @serena-ruan)
- [Models] Add support for
completions
in the OpenAI flavor (#9838, @santiagxf) - [Models] Add support for inference parameters for the OpenAI flavor (#9909, @santiagxf)
- [Models] Introduce support for configuration arguments to be specified when loading a model (#9251, @santiagxf)
- [Models] Add support for integrated Azure AD authentication for the OpenAI flavor (#9704, @santiagxf)
- [Models / Scoring] Introduce support for model training lineage in model serving (#9402, @M4nouel)
- [Model Registry] Introduce the
copy_model_version
client API for copying model versions across registered models (#9946, #10078, #10140, @jerrylian-db) - [Tracking] Expand the limits of parameter value length from 500 to 6000 (#9709, @serena-ruan)
- [Tracking] Introduce support for Spark 3.5's SparkConnect mode within MLflow to allow logging models created using this operation mode of Spark (#9534, @WeichenXu123)
- [Tracking] Add support for logging system metrics to the MLflow fluent API (#9557, #9712, #9714, @chenmoneygithub)
- [Tracking] Add callbacks within MLflow for Keras and Tensorflow (#9454, #9637, #9579, @chenmoneygithub)
- [Tracking] Introduce a fluent login API for Databricks within MLflow (#9665, #10180, @chenmoneygithub)
- [Tracking] Add support for customizing auth for http requests from the MLflow client via a plugin extension (#10049, @lu-ohai)
- [Tracking] Introduce experimental asynchronous logging support for metrics, params, and tags (#9705, @sagarsumant)
- [Auth] Modify the behavior of user creation in MLflow Authentication so that only admins can create new users (#9700, @gabrielfu)
- [Artifacts] Add support for using
xethub
as an artifact store via a plugin extension (#9957, @Kelton8Z) - [UI] Add new opt-in Model Registry UI that supports model aliases and tags (#10163, @hubertzub-db, @jerrylian-db)
Bug fixes:
- [Evaluate] Fix a bug with Azure OpenAI configuration usage within MLflow evaluate (#9982, @sunishsheth2009)
- [Models] Fix a data consistency issue when saving models that have been loaded in heterogeneous memory configuration within the transformers flavor (#10087, @BenWilson2)
- [Models] Fix an issue in the transformers flavor for complex input types by adding dynamic dataframe typing (#9044, @wamartin-aml)
- [Models] Fix an issue in the langchain flavor to provide support for chains with multiple outputs (#9497, @bbqiu)
- [Docker] Fix an issue with Docker image generation by changing the default env-manager to virtualenv (#9938, @Beramos)
- [Auth] Fix an issue with complex passwords in MLflow Auth to support a richer character set range (#9760, @dotdothu)
- [R] Fix a bug with configuration access when running MLflow R in Databricks (#10117, @zacdav-db)
Documentation updates:
- [Docs] Introduce the first phase of a larger documentation overhaul (#10197, @BenWilson2)
- [Docs] Add guide for LLM eval (#10058, #10199, @chenmoneygithub)
- [Docs] Add instructions on how to force single file serialization within the onnx flavor's save and log functions (#10178, @BenWilson2)
- [Docs] Add documentation for the relevance metric for MLflow evaluate (#10170, @sunishsheth2009)
- [Docs] Add a style guide for the contributing guide for how to structure pydoc strings (#9907, @mberk06)
- [Docs] Fix issues with the pytorch lightning autolog code example (#9964, @chenmoneygithub)
- [Docs] Update the example for
mlflow.data.from_numpy()
(#9885, @chenmoneygithub) - [Docs] Add clear instructions for installing MLflow within R (#9835, @darshan8850)
- [Docs] Update model registry documentation to add content regarding support for model aliases (#9721, @jerrylian-db)
Small bug fixes and documentation updates:
#10202, #10189, #10188, #10159, #10175, #10165, #10154, #10083, #10082, #10081, #10071, #10077, #10070, #10053, #10057, #10055, #10020, #9928, #9929, #9944, #9979, #9923, #9842, @annzhang-db; #10203, #10196, #10172, #10176, #10145, #10115, #10107, #10054, #10056, #10018, #9976, #9999, #9998, #9995, #9978, #9973, #9975, #9972, #9974, #9960, #9925, #9920, @prithvikannan; #10144, #10166, #10143, #10129, #10059, #10123, #9555, #9619, @bbqiu; #10187, #10191, #10181, #10179, #10151, #10148, #10126, #10119, #10099, #10100, #10097, #10089, #10096, #10091, #10085, #10068, #10065, #10064, #10060, #10023, #10030, #10028, #10022, #10007, #10006, #9988, #9961, #9963, #9954, #9953, #9937, #9932, #9931, #9910, #9901, #9852, #9851, #9848, #9847, #9841, #9844, #9825, #9820, #9806, #9802, #9800, #9799, #9790, #9787, #9791, #9788, #9785, #9786, #9784, #9754, #9768, #9770, #9753, #9697, #9749, #9747, #9748, #9751, #9750, #9729, #9745, #9735, #9728, #9725, #9716, #9694, #9681, #9666, #9643, #9641, #9621, #9607, @harupy; #10200, #10201, #10142, #10139, #10133, #10090, #10086, #9934, #9933, #9845, #9831, #9794, #9692, #9627, #9626, @chenmoneygithub; #10110, @wenfeiy-db; #10195, #9895, #9880, #9679, @BenWilson2; #10174, #10177, #10109, #9706, @jerrylian-db; #10113, #9765, @smurching; #10150, #10138, #10136, @dbczumar; #10153, #10032, #9986, #9874, #9727, #9707, @serena-ruan; #10155, @shaotong-db; #10160, #10131, #10048, #10024, #10017, #10016, #10002, #9966, #9924, @sunishsheth2009; #10121, #10116, #10114, #10102, #10098, @B-Step62; #10095, #10026, #9991, @daniellok-db; #10050, @Dennis40816; #10062, #9868, @Gekko0114; #10033, @Anushka-Bhowmick; #9983, #10004, #9958, #9926, #9690, @liangz1; #9997, #9940, #9922, #9919, #9890, #9888, #9889, #9810, @TomeHirata; #9994, #9970, #9950, @lightnessofbein; #9965, #9677, @ShorthillsAI; #9906, @jessechancy; #9942, #9771, @Sai-Suraj-27; #9902, @remyleone; #9892, #9865, #9866, #9853, @montanarograziano; #9875, @Raghavan-B; #9858, @Salz0; #9878, @maksboyarin; #9882, @lukasz-gawron; #9827, @Bncer; #9819, @gabrielfu; #9792, @harshk461; #9726, @Chiragasourabh; #9663, @Abhishek-TyRnT; #9670, @mberk06; #9755, @simonlsk; #9757, #9775, #9776, #9774, @AmirAflak; #9782, @garymm; #9756, @issamarabi; #9645, @shichengzhou-db; #9671, @zhe-db; #9660, @mingyu89; #9575, @akshaya-a; #9629, @pnacht; #9876, @C-K-Loan
MLflow 2.7.1 is a patch release containing the following features, bug fixes and changes:
Features:
- [Gateway / Databricks] Add the
set_limits
andget_limits
APIs for AI Gateway routes within Databricks (#9516, @zhe-db) - [Artifacts / Databricks] Add support for parallelized download and upload of artifacts within Unity Catalog (#9498, @jerrylian-db)
Bug fixes:
- [Models / R] Fix a critical bug with the
R
client that prevents models from being loaded (#9624, @BenWilson2) - [Artifacts / Databricks] Disable multi-part download functionality for UC Volumes local file destination when downloading models (#9631, @BenWilson2)
Small bug fixes and documentation updates:
#9640, @annzhang-db; #9622, @harupy
MLflow 2.7.0 includes several major features and improvements
- [UI / Gateway] We are excited to announce the Prompt Engineering UI. This new addition offers a suite of tools tailored for efficient prompt development, testing, and evaluation for LLM use cases. Integrated directly into the MLflow AI Gateway, it provides a seamless experience for designing, tracking, and deploying prompt templates. To read about this new feature, see the documentation at https://mlflow.org/docs/latest/llms/prompt-engineering.html (#9503, @prithvikannan)
Features:
- [Gateway] Introduce
MosaicML
as a supported provider for the MLflowAI Gateway
(#9459, @arpitjasa-db) - [Models] Add support for using a snapshot download location when loading a
transformers
model aspyfunc
(#9362, @serena-ruan) - [Server-infra] Introduce plugin support for MLflow
Tracking Server
authentication (#9191, @barrywhart) - [Artifacts / Model Registry] Add support for storing artifacts using the
R2
backend (#9490, @shichengzhou-db) - [Artifacts] Improve upload and download performance for Azure-based artifact stores (#9444, @jerrylian-db)
- [Sagemaker] Add support for deploying models to Sagemaker Serverless inference endpoints (#9085, @dogeplusplus)
Bug fixes:
- [Gateway] Fix a credential expiration bug by re-resolving
AI Gateway
credentials before each request (#9518, @dbczumar) - [Gateway] Fix a bug where
search_routes
would raise an exception when no routes have been defined on theAI Gateway
server (#9387, @QuentinAmbard) - [Gateway] Fix compatibility issues with
pydantic
2.x forAI gateway
(#9339, @harupy) - [Gateway] Fix an initialization issue in the
AI Gateway
that could render MLflow nonfunctional at import if dependencies were conflicting. (#9337, @BenWilson2) - [Artifacts] Fix a correctness issue when downloading large artifacts to
fuse mount
paths onDatabricks
(#9545, @BenWilson2)
Documentation updates:
- [Docs] Add documentation for the
Giskard
community plugin formlflow.evaluate
(#9183, @rabah-khalek)
Small bug fixes and documentation updates:
#9605, #9603, #9602, #9595, #9597, #9587, #9590, #9588, #9586, #9584, #9583, #9582, #9581, #9580, #9577, #9546, #9566, #9569, #9562, #9564, #9561, #9528, #9506, #9503, #9492, #9491, #9485, #9445, #9430, #9429, #9427, #9426, #9424, #9421, #9419, #9409, #9408, #9407, #9394, #9389, #9395, #9393, #9390, #9370, #9356, #9359, #9357, #9345, #9340, #9328, #9329, #9326, #9304, #9325, #9323, #9322, #9319, #9314, @harupy; #9568, #9520, @dbczumar; #9593, @jerrylian-db; #9574, #9573, #9480, #9332, #9335, @BenWilson2; #9556, @shichengzhou-db; #9570, #9540, #9533, #9517, #9354, #9453, #9338, @prithvikannan; #9565, #9560, #9536, #9504, #9476, #9481, #9450, #9466, #9418, #9397, @serena-ruan; #9489, @dnerini; #9512, #9479, #9355, #9351, #9289 @chenmoneygithub; #9488, @bbqiu; #9474, @apurva-koti; #9505, @arpitjasa-db; #9261, @donour; #9336, #9414, #9353, @mberk06; #9451, @Bncer; #9432, @barrywhart; #9347, @GraceBrigham; #9428, #9420, #9406, @WeichenXu123; #9410, @aloahPGF; #9396, #9384, #9372, @Godwin-T; #9373, @fabiansefranek; #9382, @Sai-Suraj-27; #9378, @saidattu2003; #9375, @Increshi; #9358, @smurching; #9366, #9330, @Dev-98; #9364, @Sandeep1005; #9349, #9348, @AmirAflak; #9308, @danilopeixoto; #9596, @ShorthillsAI; #9567, @Beramos; #9524, @rabah-khalek; #9312, @dependabot[bot]
MLflow 2.6.0 includes several major features and improvements
Features:
- [Models / Scoring] Add support for passing extra params during inference for PyFunc models (#9068, @serena-ruan)
- [Gateway] Add support for MLflow serving to MLflow AI Gateway (#9199, @BenWilson2)
- [Tracking] Support
save_kwargs
formlflow.log_figure
to specify extra options when saving a figure (#9179, @stroblme) - [Artifacts] Display progress bars when uploading/download artifacts (#9195, @serena-ruan)
- [Models] Add support for logging LangChain's retriever models (#8808, @liangz1)
- [Tracking] Add support to log customized tags to runs created by autologging (#9114, @thinkall)
Bug fixes:
- [Models] Fix
text_pair
functionality for transformersTextClassification
pipelines (#9215, @BenWilson2) - [Models] Fix LangChain compatibility with SQLDatabase (#9192, @dbczumar)
- [Tracking] Remove patching
sklearn.metrics.get_scorer_names
inmlflow.sklearn.autolog
to avoid duplicate logging (#9095, @WeichenXu123)
Documentation updates:
- [Docs / Examples] Add examples and documentation for MLflow AI Gateway support for MLflow model serving (#9281, @BenWilson2)
- [Docs / Examples] Add
sentence-transformers
doc & example (#9047, @es94129)
Deprecation:
- [Models] The
mlflow.mleap
module has been marked as deprecated and will be removed in a future release (#9311, @BenWilson2)
Small bug fixes and documentation updates:
#9309, #9252, #9198, #9189, #9186, #9184, @BenWilson2; #9307, @AmirAflak; #9285, #9126, @dependabot[bot]; #9302, #9209, #9194, #9187, #9175, #9177, #9163, #9161, #9129, #9123, #9053, @serena-ruan; #9305, #9303, #9271, @KekmaTime; #9300, #9299, @itsajay1029; #9294, #9293, #9274, #9268, #9264, #9246, #9255, #9253, #9254, #9245, #9202, #9243, #9238, #9234, #9233, #9227, #9226, #9223, #9224, #9222, #9225, #9220, #9208, #9212, #9207, #9203, #9201, #9200, #9154, #9146, #9147, #9153, #9148, #9145, #9136, #9132, #9131, #9128, #9121, #9124, #9125, #9108, #9103, #9100, #9098, #9101, @harupy; #9292, @Aman123lug; #9290, #9164, #9157, #9086, @Bncer; #9291, @kunal642; #9284, @NavneetSinghArora; #9286, #9262, #9142, @smurching; #9267, @tungbq; #9258, #9250, @Kunj125; #9167, #9139, #9120, #9118, #9097, @viktoriussuwandi; #9244, #9240, #9239, @Sai-Suraj-27; #9221, #9168, #9130, @gabrielfu; #9218, @tjni; #9216, @Rukiyav; #9158, #9051, @EdAbati; #9211, @scarlettrobe; #9049, @annzhang-db; #9140, @kriscon-db; #9141, @xAIdrian; #9135, @liangz1; #9067, @jmmonteiro; #9112, @WeichenXu123; #9106, @shaikmoeed; #9105, @Ankit8848; #9104, @arnabrahman
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:
- [Models] Add Support to the LangChain flavor for chains that contain unserializable components (#8736, @liangz1)
- [Scoring] Infer spark udf return type from model output schema (#8934, @WeichenXu123)
- [Models] Add support for automated signature inference (#8860, #8782 #8795, #8725, @jerrylian-db)
Bug fixes:
- [Security] Improve robustness to LFI attacks on Windows by enhancing path validation (#8999, @serena-ruan)
- If you are using
mlflow server
ormlflow ui
on Windows, we recommend upgrading to MLflow 2.5.0 as soon as possible.
- If you are using
- [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. Themlflow.gluon
module will be removed in a future release. (#8968, @harupy)
Small bug fixes and documentation updates:
#9069, #9056, #9055, #9054, #9048, #9043, #9035, #9034, #9037, #9038, #8993, #8966, #8985, @BenWilson2; #9039, #9036, #8902, #8924, #8866, #8861, #8810, #8761, #8544, @jerrylian-db; #8903, @smurching; #9080, #9079, #9078, #9076, #9075, #9074, #9071, #9063, #9062, #9032, #9031, #9027, #9023, #9022, #9020, #9005, #8994, #8979, #8983, #8984, #8982, #8970, #8962, #8969, #8968, #8959, #8960, #8958, #8956, #8955, #8954, #8949, #8950, #8952, #8948, #8946, #8947, #8943, #8944, #8916, #8917, #8933, #8929, #8932, #8927, #8930, #8925, #8921, #8873, #8915, #8909, #8908, #8911, #8910, #8907, #8906, #8898, #8893, #8889, #8892, #8891, #8887, #8875, #8876, #8882, #8874, #8868, #8872, #8869, #8828, #8852, #8857, #8853, #8854, #8848, #8850, #8840, #8835, #8832, #8831, #8830, #8829, #8839, #8833, #8838, #8819, #8814, #8825, #8818, #8787, #8775, #8749, #8766, #8756, #8753, #8751, #8748, #8744, #8731, #8717, #8730, #8691, #8720, #8723, #8719, #8688, #8721, #8715, #8716, #8718, #8696, #8698, #8692, #8693, #8690, @harupy; #9030, @AlimurtuzaCodes; #9029, #9025, #9021, #9013, @viktoriussuwandi; #9010, @Bncer; #9011, @Pecunia201; #9007, #9003, @EdAbati; #9002, @prithvikannan; #8991, #8867, @AveshCSingh; #8951, #8896, #8888, #8849, @gabrielfu; #8913, #8885, #8871, #8870, #8788, #8772, #8771, @serena-ruan; #8879, @maciejskorski; #7752, @arunkumarkota; #9083, #9081, #8765, #8742, #8685, #8682, #8683, @dbczumar; #8791, @mhattingpete; #8739, @yunpark93
MLflow 2.4.2 is a patch release containing the following bug fixes and changes:
Bug fixes:
- [Models] Add compatibility for legacy transformers serialization (#8964, @BenWilson2)
- [Models] Fix downloading MLmodel files from alias-based models:/ URIs (#8764, @smurching)
- [Models] Fix reading model flavor config from URI for models in UC (#8728, @smurching)
- [Models] Support
feature_deps
in ModelVersion creation for UC (#8867, #8815, @AveshCSingh) - [Models] Add support for listing artifacts in UC model registry artifact repo (#8803, @smurching)
- [Core] Include resources for recipes in mlflow-skinny (#8895, @harupy)
- [UI] Enable datasets tracking UI (#8886, @harupy)
- [Artifacts] Use
MLFLOW_ENABLE_MULTIPART_DOWNLOAD
inDatabricksArtifactRepository
(#8884, @harupy)
Documentation updates:
- [Examples / Docs] Add question-answering and summarization examples and docs with LLMs (#8695, @dbczumar)
- [Examples / Docs] Add johnsnowlabs flavor example and doc (#8689, @C-K-Loan)
Small bug fixes and documentation updates:
#8966, @BenWilson2; #8881, @harupy; #8846, #8760, @smurching
MLflow 2.4.1 is a patch release containing the following features, bug fixes and changes:
Features:
- [Tracking] Extend SearchRuns to support datasets (#8622, @prithvikannan)
- [Models] Add an
mlflow.johnsnowlabs
flavor for thejohnsnowlabs
package (#8556, @C-K-Loan) - [Models] Add a warning for duplicate pip requirements specified in
save_model
andlog_model
for thetransformers
flavor (#8678, @BenWilson2)
Bug fixes:
- [Security] Improve robustness to LFI attacks (#8648, @serena-ruan)
- If you are using
mlflow server
ormlflow ui
, we recommend upgrading to MLflow 2.4.1 as soon as possible.
- If you are using
- [Models] Fix an issue with
transformers
serialization for ModelCards that contain invalid characters (#8652, @BenWilson2) - [Models] Fix connection pooling deadlocks that occurred during large file downloads (#8682, @dbczumar; #8660, @harupy)
Small bug fixes and documentation updates:
#8677, #8674, #8646, #8647, @dbczumar; #8654, #8653, #8660, #8650, #8642, #8636, #8599, #8637, #8608, #8633, #8623, #8628, #8619, @harupy; #8655, #8609, @BenWilson2; #8648, @serena-ruan; #8521, @ka1mar; #8638, @smurching; #8634, @PenHsuanWang
MLflow 2.4.0 includes several major features and improvements
Features:
- [Tracking] Introduce dataset tracking APIs:
mlflow.data
andmlflow.log_input()
(#8186, @prithvikannan) - [Tracking] Add
mlflow.log_table()
andmlflow.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
andLLMChain
inmlflow.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 themlflow.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
withFlask<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)
Small bug fixes and documentation updates:
#8611, #8587, @dbczumar; #8617, #8620, #8615, #8603, #8604, #8601, #8596, #8598, #8597, #8589, #8580, #8581, #8575, #8582, #8577, #8576, #8578, #8561, #8568, #8551, #8528, #8550, #8489, #8530, #8534, #8533, #8532, #8524, #8520, #8517, #8516, #8515, #8514, #8506, #8503, #8500, #8504, #8496, #8486, #8485, #8468, #8471, #8473, #8470, #8458, #8447, #8446, #8434, @harupy; #8607, #8538, #8513, #8452, #8466, #8465, @serena-ruan; #8586, #8595, @prithvikannan; #8593, #8541, @kriscon-db; #8592, #8566, @annzhang-db; #8588, #8565, #8559, #8537, @BenWilson2; #8545, @apurva-koti; #8564, @DavidSpek; #8436, #8490, @jerrylian-db; #8505, @eliaskoromilas; #8483, @WeichenXu123; #8472, @leqiao-1; #8429, @jinzhang21; #8581, #8548, #8499, @gabrielfu;
MLflow 2.3.2 is a patch release containing the following features, bug fixes and changes:
Features:
- [Models] Add GPU support for
transformers
modelspyfunc
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 intransformers
pipelines (#8421, @BenWilson2) - [Models] Add support for
Feature Extraction
pipelines in thetransformers
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 thetransformers
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:
- [Examples] Add
signature
logging to all examples and documentation (#8410, #8401, #8400, #8387 @jerrylian-db) - [Examples] Add
sentence-transformers
examples to thetransformers
examples suite (#8425, @BenWilson2) - [Docs] Add a new MLflow Quickstart documentation page (#8171, @lobrien)
- [Docs] Add a new introduction to MLflow page (#8365, @lobrien)
- [Docs] Add a community model plugin example and documentation for
trubrics
(#8371, @jeffkayne) - [Docs] Add
gluon
pyfunc example to Model flavor documentation (#8403, @ericvincent18) - [Docs] Add
statsmodels
pyfunc example toModels
flavor documentation (#8394, @ericvincent18)
Small bug fixes and documentation updates:
#8415, #8412, #8411, #8355, #8354, #8353, #8348, @harupy; #8374, #8367, #8350, @dbczumar; #8358 @mrkaye97; #8392, #8362, @smurching; #8427, #8408, #8399, #8381, @BenWilson2; #8395, #8390, @jerrylian-db; #8402, #8398, @WeichenXu123; #8377, #8363, @arpitjasa-db; #8385, @prithvikannan; #8418, @Jeukoh;
MLflow 2.3.1 is a patch release containing the following bug fixes and changes:
Bug fixes:
- [Security] Fix critical LFI attack vulnerability by disabling the ability to provide relative paths in registered model sources (#8281, @BenWilson2)
- If you are using
mlflow server
ormlflow ui
, we recommend upgrading to MLflow 2.3.1 as soon as possible. For more details, see https://github.com/mlflow/mlflow/security/advisories/GHSA-xg73-94fp-g449.
- If you are using
- [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)
Documentation updates:
- [Docs] Add an H2O pyfunc usage example to the models documentation (#8292, @ericvincent18)
- [Examples] Add a TensorFlow Core 2.x API usage example (#8235, @dheerajnbhat)
Small bug fixes and documentation updates:
#8324, #8325, @smurching; #8313, @dipanjank; #8323, @liangz1; #8331, #8328, #8319, #8316, #8308, #8293, #8289, #8283, #8284, #8285, #8282, #8241, #8270, #8272, #8271, #8268, @harupy; #8312, #8294, #8295, #8279, #8267, @BenWilson2; #8290, @jinzhang21; #8257, @WeichenXu123; #8307, @arpitjasa-db
MLflow 2.3.0 includes several major features and improvements
Features:
- [Models] Introduce a new
transformers
named flavor (#8236, #8181, #8086, @BenWilson2) - [Models] Introduce a new
openai
named flavor (#8191, #8155, @harupy) - [Models] Introduce a new
langchain
named flavor (#8251, #8197, @liangz1, @sunishsheth2009) - [Models] Add support for
Pytorch
andLightning
2.0 (#8072, @shrinath-suresh) - [Tracking] Add support for logging LLM input, output, and prompt artifacts (#8234, #8204, @sunishsheth2009)
- [Tracking] Add support for HTTP Basic Auth in the MLflow tracking server (#8130, @gabrielfu)
- [Tracking] Add
search_model_versions
to the fluent API (#8223, @mariusschlegel) - [Artifacts] Add support for parallelized artifact downloads (#8116, @apurva-koti)
- [Artifacts] Add support for parallelized artifact uploads for AWS (#8003, @harupy)
- [Artifacts] Add content type headers to artifact upload requests for the
HttpArtifactRepository
(#8048, @WillEngler) - [Model Registry] Add alias support for logged models within Model Registry (#8164, #8094, #8055 @arpitjasa-db)
- [UI] Add support for custom domain git providers (#7933, @gusghrlrl101)
- [Scoring] Add plugin support for customization of MLflow serving endpoints (#7757, @jmahlik)
- [Scoring] Add support to MLflow serving that allows configuration of multiple inference workers (#8035, @M4nouel)
- [Sagemaker] Add support for asynchronous inference configuration on Sagemaker (#8009, @thomasbell1985)
- [Build] Remove
shap
as a core dependency of MLflow (#8199, @jmahlik)
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)
Small bug fixes and documentation updates:
#8262, #8252, #8250, #8228, #8221, #8203, #8134, #8040, #7994, #7934, @BenWilson2; #8258, #8255, #8253, #8248, #8247, #8245, #8243, #8246, #8244, #8242, #8240, #8229, #8198, #8192, #8112, #8165, #8158, #8152, #8148, #8144, #8143, #8120, #8107, #8105, #8102, #8088, #8089, #8096, #8075, #8073, #8076, #8063, #8064, #8033, #8024, #8023, #8021, #8015, #8005, #7982, #8002, #7987, #7981, #7968, #7931, #7930, #7929, #7917, #7918, #7916, #7914, #7913, @harupy; #7955, @arjundc-db; #8219, #8110, #8093, #8087, #8091, #8092, #8029, #8028, #8031, @jerrylian-db; #8187, @apurva-koti; #8210, #8001, #8000, @arpitjasa-db; #8161, #8127, #8095, #8090, #8068, #8043, #7940, #7924, #7923, @dbczumar; #8147, @morelen17; #8106, @WeichenXu123; #8117, @eltociear; #8100, @laerciop; #8080, @elado; #8070, @grofte; #8066, @yukimori; #8027, #7998, @liangz1; #7999, @martlaf; #7964, @viditjain99; #7928, @alekseyolg; #7909, #7901, #7844, @smurching; #7971, @n30111; #8012, @mingyu89; #8137, @lobrien; #7992, @robmarkcole; #8263, @sunishsheth2009
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 arun_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 inFileStore.get_registered_model
(#7965, @harupy)
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) - [Model Registry] Patch for GHSA-xg73-94fp-g449 (#7908, @harupy)
- [Model Registry] Patch for GHSA-wp72-7hj9-5265 (#7965, @harupy)
MLflow 2.2.0 includes several major features and improvements
Features:
- [Recipes] Add support for score calibration to the classification recipe (#7744, @sunishsheth2009)
- [Recipes] Add automatic label encoding to the classification recipe (#7711, @sunishsheth2009)
- [Recipes] Support custom data splitting logic in the classification and regression recipes (#7815, #7588, @sunishsheth2009)
- [Recipes] Introduce customizable MLflow Run name prefixes to the classification and regression recipes (#7746, @kamalesh0406; #7763, @sunishsheth2009)
- [UI] Add a new Chart View to the MLflow Experiment Page for model performance insights (#7864, @hubertzub-db, @apurva-koti, @prithvikannan, @ridhimag11, @sunishseth2009, @dbczumar)
- [UI] Modernize and improve parallel coordinates chart for model tuning (#7864, @hubertzub-db, @apurva-koti, @prithvikannan, @ridhimag11, @sunishseth2009, @dbczumar)
- [UI] Add typeahead suggestions to the MLflow Experiment Page search bar (#7864, @hubertzub-db, @apurva-koti, @prithvikannan, @ridhimag11, @sunishseth2009, @dbczumar)
- [UI] Improve performance of Experiments Sidebar for large numbers of experiments (#7804, @jmahlik)
- [Tracking] Introduce autologging support for native PyTorch models (#7627, @temporaer)
- [Tracking] Allow specifying
model_format
when autologging XGBoost models (#7781, @guyrosin) - [Tracking] Add
MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT
environment variable to configure artifact operation timeouts (#7783, @wamartin-aml) - [Artifacts] Include
Content-Type
response headers for artifacts downloaded frommlflow server
(#7827, @bali0019) - [Model Registry] Introduce the
searchModelVersions()
API to the Java client (#7880, @gabrielfu) - [Model Registry] Introduce
max_results
,order_by
andpage_token
arguments toMlflowClient.search_model_versions()
(#7623, @serena-ruan) - [Models] Support logging large ONNX models by using external data (#7808, @dogeplusplus)
- [Models] Add support for logging Diviner models fit in Spark (#7800, @BenWilson2)
- [Models] Introduce
MLFLOW_DEFAULT_PREDICTION_DEVICE
environment variable to set the device for pyfunc model inference (#7922, @ankit-db) - [Scoring] Publish official Docker images for the MLflow Model scoring server at github.com/mlflow/mlflow/pkgs (#7759, @dbczumar)
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
andartifact_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:
- [Docs] Add an example of saving and loading Spark MLlib models with MLflow (#7706, @dipanjank)
- [Docs] Add usage examples for
mlflow.lightgbm
APIs (#7565, @canerturkseven) - [Docs] Add an example of custom model flavor creation with
sktime
(#7624, @benjaminbluhm) - [Docs] Clarify
precision_recall_auc
metric calculation inmlflow.evaluate()
(#7701, @BenWilson2) - [Docs] Remove outdated example links (#7587, @asloan7)
Small bug fixes and documentation updates:
#7866, #7751, #7724, #7699, #7697, #7666, @alekseyolg; #7896, #7861, #7858, #7862, #7872, #7859, #7863, #7767, #7766, #7765, #7741, @smurching; #7895, #7877, @viditjain99; #7898, @midhun1998; #7891, #7892, #7886, #7882, #7883, #7875, #7874, #7871, #7868, #7854, #7847, #7845, #7838, #7830, #7837, #7836, #7834, #7831, #7828, #7825, #7826, #7824, #7823, #7778, #7780, #7776, #7775, #7773, #7772, #7769, #7756, #7768, #7764, #7685, #7726, #7722, #7720, #7423, #7712, #7710, #7713, #7688, #7663, #7674, #7673, #7672, #7662, #7653, #7646, #7615, #7614, #7586, #7601, #7598, #7602, #7599, #7577, #7585, #7583, #7584, @harupy; #7865, #7803, #7753, #7719, @dipanjank; #7796, @serena-ruan; #7849, @turbotimon; #7822, #7600, @WeichenXu123; #7811, @guyrosin; #7812, #7788, #7787, #7748, #7730, #7616, #7593, @dbczumar; #7793, @Joel-hanson; #7792, #7694, #7643, @BenWilson2; #7771, #7657, #7644, @nsenno-dbr; #7738, @wkrt7; #7740, @Ark-kun; #7739, #7733, @bali0019; #7723, @andrehp; #7691, #7582, @agoyot; #7721, @Eseeldur; #7709, @srowen; #7693, @ry3s; #7649, @funkypenguin; #7665, @benjaminbluhm; #7668, @eltociear; #7550, @danielhstahl; #7920, @arjundc-db
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 tomlflow 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 andlog_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
andconda.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 withmlflow.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)
Small bug fixes and documentation updates:
#7571, #7543, #7529, #7435, #7399, @WeichenXu123; #7568, @xiaoye-hua; #7549, #7557, #7509, #7498, #7499, #7485, #7486, #7484, #7391, #7388, #7390, #7381, #7366, #7348, #7346, #7334, #7340, #7323, @BenWilson2; #7561, #7562, #7560, #7553, #7546, #7539, #7544, #7542, #7541, #7533, #7507, #7470, #7469, #7467, #7466, #7464, #7453, #7449, #7450, #7440, #7430, #7436, #7429, #7426, #7410, #7406, #7409, #7407, #7405, #7396, #7393, #7395, #7384, #7376, #7379, #7375, #7354, #7353, #7351, #7352, #7350, #7345, #6493, #7343, #7344, @harupy; #7494, @dependabot[bot]; #7526, @tobycheese; #7489, @liangz1; #7534, @Jingnan-Jia; #7496, @danielhstahl; #7504, #7503, #7459, #7454, #7447, @tsugumi-sys; #7461, @wkrt7; #7451, #7414, #7372, #7289, @sunishsheth2009; #7441, @ikrizanic; #7432, @Pochingto; #7386, @jhallard; #7370, #7373, #7371, #7336, #7341, #7342, @dbczumar; #7335, @prithvikannan
The 2.0.1 version of MLflow is a major milestone release that focuses on simplifying the management of end-to-end MLOps workflows, providing new feature-rich functionality, and expanding upon the production-ready MLOps capabilities offered by MLflow. This release contains several important breaking changes from the 1.x API, additional major features and improvements.
Features:
- [Recipes] MLflow Pipelines is now MLflow Recipes - a framework that enables data scientists to quickly develop high-quality models and deploy them to production
- [Recipes] Add support for classification models to MLflow Recipes (#7082, @bbarnes52)
- [UI] Introduce support for pinning runs within the experiments UI (#7177, @harupy)
- [UI] Simplify the layout and provide customized displays of metrics, parameters, and tags within the experiments UI (#7177, @harupy)
- [UI] Simplify run filtering and ordering of runs within the experiments UI (#7177, @harupy)
- [Tracking] Update
mlflow.pyfunc.get_model_dependencies()
to download all referenced requirements files for specified models (#6733, @harupy) - [Tracking] Add support for selecting the Keras model
save_format
used bymlflow.tensorflow.autolog()
(#7123, @balvisio) - [Models] Set
mlflow.evaluate()
status to stable as it is now a production-ready API - [Models] Simplify APIs for specifying custom metrics and custom artifacts during model evaluation with
mlflow.evaluate()
(#7142, @harupy) - [Models] Correctly infer the positive label for binary classification within
mlflow.evaluate()
(#7149, @dbczumar) - [Models] Enable automated signature logging for
tensorflow
andkeras
models whenmlflow.tensorflow.autolog()
is enabled (#6678, @BenWilson2) - [Models] Add support for native Keras and Tensorflow Core models within
mlflow.tensorflow
(#6530, @WeichenXu123) - [Models] Add support for defining the
model_format
used bymlflow.xgboost.save/log_model()
(#7068, @AvikantSrivastava) - [Scoring] Overhaul the model scoring REST API to introduce format indicators for inputs and support multiple output fields (#6575, @tomasatdatabricks; #7254, @adriangonz)
- [Scoring] Add support for ragged arrays in model signatures (#7135, @trangevi)
- [Java] Add
getModelVersion
API to the java client (#6955, @wgottschalk)
Breaking Changes:
The following list of breaking changes are arranged by their order of significance within each category.
- [Core] Support for Python 3.7 has been dropped. MLflow now requires Python >=3.8
- [Recipes]
mlflow.pipelines
APIs have been replaced withmlflow.recipes
- [Tracking / Registry] Remove
/preview
routes for Tracking and Model Registry REST APIs (#6667, @harupy) - [Tracking] Remove deprecated
list
APIs for experiments, models, and runs from Python, Java, R, and REST APIs (#6785, #6786, #6787, #6788, #6800, #6868, @dbczumar) - [Tracking] Remove deprecated
runs
response field fromGet Experiment
REST API response (#6541, #6524 @dbczumar) - [Tracking] Remove deprecated
MlflowClient.download_artifacts
API (#6537, @WeichenXu123) - [Tracking] Change the behavior of environment variable handling for
MLFLOW_EXPERIMENT_NAME
such that the value is always used when creating an experiment (#6674, @BenWilson2) - [Tracking] Update
mlflow server
to run in--serve-artifacts
mode by default (#6502, @harupy) - [Tracking] Update Experiment ID generation for the Filestore backend to enable threadsafe concurrency (#7070, @BenWilson2)
- [Tracking] Remove
dataset_name
andon_data_{name | hash}
suffixes frommlflow.evaluate()
metric keys (#7042, @harupy) - [Models / Scoring / Projects] Change default environment manager to
virtualenv
instead ofconda
for model inference and project execution (#6459, #6489 @harupy) - [Models] Move Keras model logging APIs to the
mlflow.tensorflow
flavor and drop support for TensorFlow Estimators (#6530, @WeichenXu123) - [Models] Remove deprecated
mlflow.sklearn.eval_and_log_metrics()
API in favor ofmlflow.evaluate()
API (#6520, @dbczumar) - [Models] Require
mlflow.evaluate()
model inputs to be specified as URIs (#6670, @harupy) - [Models] Drop support for returning custom metrics and artifacts from the same function when using
mlflow.evaluate()
, in favor ofcustom_artifacts
(#7142, @harupy) - [Models] Extend
PyFuncModel
spec to supportconda
andvirtualenv
subfields (#6684, @harupy) - [Scoring] Remove support for defining input formats using the
Content-Type
header (#6575, @tomasatdatabricks; #7254, @adriangonz) - [Scoring] Replace the
--no-conda
CLI option argument for native serving with--env-manager='local'
(#6501, @harupy) - [Scoring] Remove public APIs for
mlflow.sagemaker.deploy()
andmlflow.sagemaker.delete()
in favor of MLflow deployments APIs, such asmlflow deployments -t sagemaker
(#6650, @dbczumar) - [Scoring] Rename input argument
df
toinputs
inmlflow.deployments.predict()
method (#6681, @BenWilson2) - [Projects] Replace the
use_conda
argument with theenv_manager
argument within therun
CLI command for MLflow Projects (#6654, @harupy) - [Projects] Modify the MLflow Projects docker image build options by renaming
--skip-image-build
to--build-image
with a default ofFalse
(#7011, @harupy) - [Integrations/Azure] Remove deprecated
mlflow.azureml
modules from MLflow in favor of theazure-mlflow
deployment plugin (#6691, @BenWilson2) - [R] Remove conda integration with the R client (#6638, @harupy)
Bug fixes:
- [Recipes] Fix rendering issue with profile cards polyfill (#7154, @hubertzub-db)
- [Tracking] Set the MLflow Run name correctly when specified as part of the
tags
argument tomlflow.start_run()
(#7228, @Cokral) - [Tracking] Fix an issue with conflicting MLflow Run name assignment if the
mlflow.runName
tag is set (#7138, @harupy) - [Scoring] Fix incorrect payload constructor error in SageMaker deployment client
predict()
API (#7193, @dbczumar) - [Scoring] Fix an issue where
DataCaptureConfig
information was not preserved when updating a Sagemaker deployment (#7281, @harupy)
Small bug fixes and documentation updates:
#7309, #7314, #7288, #7276, #7244, #7207, #7175, #7107, @sunishsheth2009; #7261, #7313, #7311, #7249, #7278, #7260, #7284, #7283, #7263, #7266, #7264, #7267, #7265, #7250, #7259, #7247, #7242, #7143, #7214, #7226, #7230, #7227, #7229, #7225, #7224, #7223, #7210, #7192, #7197, #7196, #7204, #7198, #7191, #7189, #7184, #7182, #7170, #7183, #7131, #7165, #7151, #7164, #7168, #7150, #7128, #7028, #7118, #7117, #7102, #7072, #7103, #7101, #7100, #7099, #7098, #7041, #7040, #6978, #6768, #6719, #6669, #6658, #6656, #6655, #6538, #6507, #6504 @harupy; #7310, #7308, #7300, #7290, #7239, #7220, #7127, #7091, #6713 @BenWilson2; #7332, #7299, #7271, #7209, #7180, #7179, #7158, #7147, #7114, @prithvikannan; #7275, #7245, #7134, #7059, @jinzhang21; #7306, #7298, #7287, #7272, #7258, #7236, @ayushthe1; #7279, @tk1012; #7219, @rddefauw; #7333, #7218, #7208, #7188, #7190, #7176, #7137, #7136, #7130, #7124, #7079, #7052, #6541 @dbczumar; #6640, @WeichenXu123; #7200, @hubertzub-db; #7121, @Gonmeso; #6988, @alonisser; #7141, @pdifranc; #7086, @jerrylian-db; #7286, @shogohida
MLflow 1.30.0 includes several major features and improvements
Features:
- [Pipelines] Introduce hyperparameter tuning support to MLflow Pipelines (#6859, @prithvikannan)
- [Pipelines] Introduce support for prediction outlier comparison to training data set (#6991, @jinzhang21)
- [Pipelines] Introduce support for recording all training parameters for reproducibility (#7026, #7094, @prithvikannan)
- [Pipelines] Add support for
Delta
tables as a datasource in the ingest step (#7010, @sunishsheth2009) - [Pipelines] Add expanded support for data profiling up to 10,000 columns (#7035, @prithvikanna)
- [Pipelines] Add support for AutoML in MLflow Pipelines using FLAML (#6959, @mshtelma)
- [Pipelines] Add support for simplified transform step execution by allowing for unspecified configuration (#6909, @apurva-koti)
- [Pipelines] Introduce a data preview tab to the transform step card (#7033, @prithvikannan)
- [Tracking] Introduce
run_name
attribute forcreate_run
,get_run
andupdate_run
APIs (#6782, #6798 @apurva-koti) - [Tracking] Add support for searching by
creation_time
andlast_update_time
for thesearch_experiments
API (#6979, @harupy) - [Tracking] Add support for search terms
run_id IN
andrun ID NOT IN
for thesearch_runs
API (#6945, @harupy) - [Tracking] Add support for searching by
user_id
andend_time
for thesearch_runs
API (#6881, #6880 @subramaniam02) - [Tracking] Add support for searching by
run_name
andrun_id
for thesearch_runs
API (#6899, @harupy; #6952, @alexacole) - [Tracking] Add support for synchronizing run
name
attribute andmlflow.runName
tag (#6971, @BenWilson2) - [Tracking] Add support for signed tracking server requests using AWSSigv4 and AWS IAM (#7044, @pdifranc)
- [Tracking] Introduce the
update_run()
API for modifying thestatus
andname
attributes of existing runs (#7013, @gabrielfu) - [Tracking] Add support for experiment deletion in the
mlflow gc
cli API (#6977, @shaikmoeed) - [Models] Add support for environment restoration in the
evaluate()
API (#6728, @jerrylian-db) - [Models] Remove restrictions on binary classification labels in the
evaluate()
API (#7077, @dbczumar) - [Scoring] Add support for
BooleanType
tomlflow.pyfunc.spark_udf()
(#6913, @BenWilson2) - [SQLAlchemy] Add support for configurable
Pool
class options forSqlAlchemyStore
(#6883, @mingyu89)
Bug fixes:
- [Pipelines] Enable Pipeline subprocess commands to create a new
SparkSession
if one does not exist (#6846, @prithvikannan) - [Pipelines] Fix a rendering issue with
bool
column types in Step Card data profiles (#6907, @sunishsheth2009) - [Pipelines] Add validation and an exception if required step files are missing (#7067, @mingyu89)
- [Pipelines] Change step configuration validation to only be performed during runtime execution of a step (#6967, @prithvikannan)
- [Tracking] Fix infinite recursion bug when inferring the model schema in
mlflow.pyspark.ml.autolog()
(#6831, @harupy) - [UI] Remove the browser error notification when failing to fetch artifacts (#7001, @kevingreer)
- [Models] Allow
mlflow-skinny
package to serve as base requirement inMLmodel
requirements (#6974, @BenWilson2) - [Models] Fix an issue with code path resolution for loading SparkML models (#6968, @dbczumar)
- [Models] Fix an issue with dependency inference in logging SparkML models (#6912, @BenWilson2)
- [Models] Fix an issue involving potential duplicate downloads for SparkML models (#6903, @serena-ruan)
- [Models] Add missing
pos_label
tosklearn.metrics.precision_recall_curve
inmlflow.evaluate()
(#6854, @dbczumar) - [SQLAlchemy] Fix a bug in
SqlAlchemyStore
whereset_tag()
updates the incorrect tags (#7027, @gabrielfu)
Documentation updates:
- [Models] Update details regarding the default
Keras
serialization format (#7022, @balvisio)
Small bug fixes and documentation updates:
#7093, #7095, #7092, #7064, #7049, #6921, #6920, #6940, #6926, #6923, #6862, @jerrylian-db; #6946, #6954, #6938, @mingyu89; #7047, #7087, #7056, #6936, #6925, #6892, #6860, #6828, @sunishsheth2009; #7061, #7058, #7098, #7071, #7073, #7057, #7038, #7029, #6918, #6993, #6944, #6976, #6960, #6933, #6943, #6941, #6900, #6901, #6898, #6890, #6888, #6886, #6887, #6885, #6884, #6849, #6835, #6834, @harupy; #7094, #7065, #7053, #7026, #7034, #7021, #7020, #6999, #6998, #6996, #6990, #6989, #6934, #6924, #6896, #6895, #6876, #6875, #6861, @prithvikannan; #7081, #7030, #7031, #6965, #6750, @bbarnes52; #7080, #7069, #7051, #7039, #7012, #7004, @dbczumar; #7054, @jinzhang21; #7055, #7037, #7036, #6949, #6951, @apurva-koti; #6815, @michaguenther; #6897, @chaturvedakash; #7025, #6981, #6950, #6948, #6937, #6829, #6830, @BenWilson2; #6982, @vadim; #6985, #6927, @kriscon-db; #6917, #6919, #6872, #6855, @WeichenXu123; #6980, @utkarsh867; #6973, #6935, @wentinghu; #6930, @mingyangge-db; #6956, @RohanBha1; #6916, @av-maslov; #6824, @shrinath-suresh; #6732, @oojo12; #6807, @ikrizanic; #7066, @subramaniam20jan; #7043, @AvikantSrivastava; #6879, @jspablo
MLflow 1.29.0 includes several major features and improvements
Features:
- [Pipelines] Improve performance and fidelity of dataset profiling in the scikit-learn regression Pipeline (#6792, @sunishsheth2009)
- [Pipelines] Add an
mlflow pipelines get-artifact
CLI for retrieving Pipeline artifacts (#6517, @prithvikannan) - [Pipelines] Introduce an option for skipping dataset profiling to the scikit-learn regression Pipeline (#6456, @apurva-koti)
- [Pipelines / UI] Display an
mlflow pipelines
CLI command for reproducing a Pipeline run in the MLflow UI (#6376, @hubertzub-db) - [Tracking] Automatically generate friendly names for Runs if not supplied by the user (#6736, @BenWilson2)
- [Tracking] Add
load_text()
,load_image()
andload_dict()
fluent APIs for convenient artifact loading (#6475, @subramaniam02) - [Tracking] Add
creation_time
andlast_update_time
attributes to the Experiment class (#6756, @subramaniam02) - [Tracking] Add official MLflow Tracking Server Dockerfiles to the MLflow repository (#6731, @oojo12)
- [Tracking] Add
searchExperiments
API to Java client and deprecatelistExperiments
(#6561, @dbczumar) - [Tracking] Add
mlflow_search_experiments
API to R client and deprecatemlflow_list_experiments
(#6576, @dbczumar) - [UI] Make URLs clickable in the MLflow Tracking UI (#6526, @marijncv)
- [UI] Introduce support for csv data preview within the artifact viewer pane (#6567, @nnethery)
- [Model Registry / Models] Introduce
mlflow.models.add_libraries_to_model()
API for adding libraries to an MLflow Model (#6586, @arjundc-db) - [Models] Add model validation support to
mlflow.evaluate()
(#6582, @jerrylian-db) - [Models] Introduce
sample_weights
support tomlflow.evaluate()
(#6806, @dbczumar) - [Models] Add
pos_label
support tomlflow.evaluate()
for identifying the positive class (#6696, @harupy) - [Models] Make the metric name prefix and dataset info configurable in
mlflow.evaluate()
(#6593, @dbczumar) - [Models] Add utility for validating the compatibility of a dataset with a model signature (#6494, @serena-ruan)
- [Models] Add
predict_proba()
support to the pyfunc representation of scikit-learn models (#6631, @skylarbpayne) - [Models] Add support for Decimal type inference to MLflow Model schemas (#6600, @shitaoli-db)
- [Models] Add new CLI command for generating Dockerfiles for model serving (#6591, @anuarkaliyev23)
- [Scoring] Add
/health
endpoint to scoring server (#6574, @gabriel-milan) - [Scoring] Support specifying a
variant_name
during Sagemaker deployment (#6486, @nfarley-soaren) - [Scoring] Support specifying a
data_capture_config
during SageMaker deployment (#6423, @jonwiggins)
Bug fixes:
- [Tracking] Make Run and Experiment deletion and restoration idempotent (#6641, @dbczumar)
- [UI] Fix an alignment bug affecting the Experiments list in the MLflow UI (#6569, @sunishsheth2009)
- [Models] Fix a regression in the directory path structure of logged Spark Models that occurred in MLflow 1.28.0 (#6683, @gwy1995)
- [Models] No longer reload the
__main__
module when loading model code (#6647, @Jooakim) - [Artifacts] Fix an
mlflow server
compatibility issue with HDFS when running in--serve-artifacts
mode (#6482, @shidianshifen) - [Scoring] Fix an inference failure with 1-dimensional tensor inputs in TensorFlow and Keras (#6796, @LiamConnell)
Documentation updates:
- [Tracking] Mark the SearchExperiments API as stable (#6551, @dbczumar)
- [Tracking / Model Registry] Deprecate the ListExperiments, ListRegisteredModels, and
list_run_infos()
APIs (#6550, @dbczumar) - [Scoring] Deprecate
mlflow.sagemaker.deploy()
in favor ofSageMakerDeploymentClient.create()
(#6651, @dbczumar)
Small bug fixes and documentation updates:
#6803, #6804, #6801, #6791, #6772, #6745, #6762, #6760, #6761, #6741, #6725, #6720, #6666, #6708, #6717, #6704, #6711, #6710, #6706, #6699, #6700, #6702, #6701, #6685, #6664, #6644, #6653, #6629, #6639, #6624, #6565, #6558, #6557, #6552, #6549, #6534, #6533, #6516, #6514, #6506, #6509, #6505, #6492, #6490, #6478, #6481, #6464, #6463, #6460, #6461, @harupy; #6810, #6809, #6727, #6648, @BenWilson2; #6808, #6766, #6729, @jerrylian-db; #6781, #6694, @marijncv; #6580, #6661, @bbarnes52; #6778, #6687, #6623, @shraddhafalane; #6662, #6737, #6612, #6595, @sunishsheth2009; #6777, @aviralsharma07; #6665, #6743, #6573, @liangz1; #6784, @apurva-koti; #6753, #6751, @mingyu89; #6690, #6455, #6484, @kriscon-db; #6465, #6689, @hubertzub-db; #6721, @WeichenXu123; #6722, #6718, #6668, #6663, #6621, #6547, #6508, #6474, #6452, @dbczumar; #6555, #6584, #6543, #6542, #6521, @dsgibbons; #6634, #6596, #6563, #6495, @prithvikannan; #6571, @smurching; #6630, #6483, @serena-ruan; #6642, @thinkall; #6614, #6597, @jinzhang21; #6457, @cnphil; #6570, #6559, @kumaryogesh17; #6560, #6540, @iamthen0ise; #6544, @Monkero; #6438, @ahlag; #3292, @dolfinus; #6637, @ninabacc-db; #6632, @arpitjasa-db
MLflow 1.28.0 includes several major features and improvements:
Features:
- [Pipelines] Log the full Pipeline runtime configuration to MLflow Tracking during Pipeline execution (#6359, @jinzhang21)
- [Pipelines] Add
pipeline.yaml
configurations to specify the Model Registry backend used for model registration (#6284, @sunishsheth2009) - [Pipelines] Support optionally skipping the
transform
step of the scikit-learn regression pipeline (#6362, @sunishsheth2009) - [Pipelines] Add UI links to Runs and Models in Pipeline Step Cards on Databricks (#6294, @dbczumar)
- [Tracking] Introduce
mlflow.search_experiments()
API for searching experiments by name and by tags (#6333, @WeichenXu123; #6227, #6172, #6154, @harupy) - [Tracking] Increase the maximum parameter value length supported by File and SQL backends to 500 characters (#6358, @johnyNJ)
- [Tracking] Introduce an
--older-than
flag tomlflow gc
for removing runs based on deletion time (#6354, @Jason-CKY) - [Tracking] Add
MLFLOW_SQLALCHEMYSTORE_POOL_RECYCLE
environment variable for recycling SQLAlchemy connections (#6344, @postrational) - [UI] Display deeply nested runs in the Runs Table on the Experiment Page (#6065, @tospe)
- [UI] Add box plot visualization for metrics to the Compare Runs page (#6308, @ahlag)
- [UI] Display tags on the Compare Runs page (#6164, @CaioCavalcanti)
- [UI] Use scientific notation for axes when viewing metric plots in log scale (#6176, @RajezMariner)
- [UI] Add button to Metrics page for downloading metrics as CSV (#6048, @rafaelvp-db)
- [UI] Include NaN and +/- infinity values in plots on the Metrics page (#6422, @hubertzub-db)
- [Tracking / Model Registry] Introduce environment variables to control retry behavior and timeouts for REST API requests (#5745, @peterdhansen)
- [Tracking / Model Registry] Make
MlflowClient
importable asmlflow.MlflowClient
(#6085, @subramaniam02) - [Model Registry] Add support for searching registered models and model versions by tags (#6413, #6411, #6320, @WeichenXu123)
- [Model Registry] Add
stage
parameter toset_model_version_tag()
(#6185, @subramaniam02) - [Model Registry] Add
--registry-store-uri
flag tomlflow server
for specifying the Model Registry backend URI (#6142, @Secbone) - [Models] Improve performance of Spark Model logging on Databricks (#6282, @bbarnes52)
- [Models] Include Pandas Series names in inferred model schemas (#6361, @RynoXLI)
- [Scoring] Make
model_uri
optional inmlflow models build-docker
to support building generic model serving images (#6302, @harupy) - [R] Support logging of NA and NaN parameter values (#6263, @nathaneastwood)
Bug fixes and documentation updates:
- [Pipelines] Improve scikit-learn regression pipeline latency by limiting dataset profiling to the first 100 columns (#6297, @sunishsheth2009)
- [Pipelines] Use
xdg-open
instead ofopen
for viewing Pipeline results on Linux systems (#6326, @strangiato) - [Pipelines] Fix a bug that skipped Step Card rendering in Jupyter Notebooks (#6378, @apurva-koti)
- [Tracking] Use the 401 HTTP response code in authorization failure REST API responses, instead of 500 (#6106, @balvisio)
- [Tracking] Correctly classify artifacts as files and directories when using Azure Blob Storage (#6237, @nerdinand)
- [Tracking] Fix a bug in the File backend that caused run metadata to be lost in the event of a failed write (#6388, @dbczumar)
- [Tracking] Adjust
mlflow.pyspark.ml.autolog()
to only log model signatures for supported input / output data types (#6365, @harupy) - [Tracking] Adjust
mlflow.tensorflow.autolog()
to log TensorFlow early stopping callback info whenlog_models=False
is specified (#6170, @WeichenXu123) - [Tracking] Fix signature and input example logging errors in
mlflow.sklearn.autolog()
for models containing transformers (#6230, @dbczumar) - [Tracking] Fix a failure in
mlflow gc
that occurred when removing a run whose artifacts had been previously deleted (#6165, @dbczumar) - [Tracking] Add missing
sqlparse
library to MLflow Skinny client, which is required for search support (#6174, @dbczumar) - [Tracking / Model Registry] Fix an
mlflow server
bug that rejected parameters and tags with empty string values (#6179, @dbczumar) - [Model Registry] Fix a failure preventing model version schemas from being downloaded with
--serve-arifacts
enabled (#6355, @abbas123456) - [Scoring] Patch the Java Model Server to support MLflow Models logged on recent versions of the Databricks Runtime (#6337, @dbczumar)
- [Scoring] Verify that either the deployment name or endpoint is specified when invoking the
mlflow deployments predict
CLI (#6323, @dbczumar) - [Scoring] Properly encode datetime columns when performing batch inference with
mlflow.pyfunc.spark_udf()
(#6244, @harupy) - [Projects] Fix an issue where local directory paths were misclassified as Git URIs when running Projects (#6218, @ElefHead)
- [R] Fix metric logging behavior for +/- infinity values (#6271, @nathaneastwood)
- [Docs] Move Python API docs for
MlflowClient
frommlflow.tracking
tomlflow.client
(#6405, @dbczumar) - [Docs] Document that MLflow Pipelines requires Make (#6216, @dbczumar)
- [Docs] Improve documentation for developing and testing MLflow JS changes in
CONTRIBUTING.rst
(#6330, @ahlag)
Small bug fixes and doc updates (#6322, #6321, #6213, @KarthikKothareddy; #6409, #6408, #6396, #6402, #6399, #6398, #6397, #6390, #6381, #6386, #6385, #6373, #6375, #6380, #6374, #6372, #6363, #6353, #6352, #6350, #6351, #6349, #6347, #6287, #6341, #6342, #6340, #6338, #6319, #6314, #6316, #6317, #6318, #6315, #6313, #6311, #6300, #6292, #6291, #6289, #6290, #6278, #6279, #6276, #6272, #6252, #6243, #6250, #6242, #6241, #6240, #6224, #6220, #6208, #6219, #6207, #6171, #6206, #6199, #6196, #6191, #6190, #6175, #6167, #6161, #6160, #6153, @harupy; #6193, @jwgwalton; #6304, #6239, #6234, #6229, @sunishsheth2009; #6258, @xanderwebs; #6106, @balvisio; #6303, @bbarnes52; #6117, @wenfeiy-db; #6389, #6214, @apurva-koti; #6412, #6420, #6277, #6266, #6260, #6148, @WeichenXu123; #6120, @ameya-parab; #6281, @nathaneastwood; #6426, #6415, #6417, #6418, #6257, #6182, #6157, @dbczumar; #6189, @shrinath-suresh; #6309, @SamirPS; #5897, @temporaer; #6251, @herrmann; #6198, @sniafas; #6368, #6158, @jinzhang21; #6236, @subramaniam02; #6036, @serena-ruan; #6430, @ninabacc-db)
MLflow 1.27.0 includes several major features and improvements:
-
[Pipelines] With MLflow 1.27.0, we are excited to announce the release of MLflow Pipelines, an opinionated framework for structuring MLOps workflows that simplifies and standardizes machine learning application development and productionization. MLflow Pipelines makes it easy for data scientists to follow best practices for creating production-ready ML deliverables, allowing them to focus on developing excellent models. MLflow Pipelines also enables ML engineers and DevOps teams to seamlessly deploy models to production and incorporate them into applications. To get started with MLflow Pipelines, check out the docs at https://mlflow.org/docs/latest/pipelines.html. (#6115)
-
[UI] Introduce UI support for searching and comparing runs across multiple Experiments (#5971, @r3stl355)
More features:
- [Tracking] When using batch logging APIs, automatically split large sets of metrics, tags, and params into multiple requests (#6052, @nzw0301)
- [Tracking] When an Experiment is deleted, SQL-based backends also move the associate Runs to the "deleted" lifecycle stage (#6064, @AdityaIyengar27)
- [Tracking] Add support for logging single-element
ndarray
and tensor instances as metrics via themlflow.log_metric()
API (#5756, @ntakouris) - [Models] Add support for
CatBoostRanker
models to themlflow.catboost
flavor (#6032, @danielgafni) - [Models] Integrate SHAP's
KernelExplainer
withmlflow.evaluate()
, enabling model explanations on categorical data (#6044, #5920, @WeichenXu123) - [Models] Extend
mlflow.evaluate()
to automatically log thescore()
outputs of scikit-learn models as metrics (#5935, #5903, @WeichenXu123)
Bug fixes and documentation updates:
- [UI] Fix broken model links in the Runs table on the MLflow Experiment Page (#6014, @hctpbl)
- [Tracking/Installation] Require
sqlalchemy>=1.4.0
upon MLflow installation, which is necessary for usage of SQL-based MLflow Tracking backends (#6024, @sniafas) - [Tracking] Fix a regression that caused
mlflow server
to rejectLogParam
API requests containing empty string values (#6031, @harupy) - [Tracking] Fix a failure in scikit-learn autologging that occurred when
matplotlib
was not installed on the host system (#5995, @fa9r) - [Tracking] Fix a failure in TensorFlow autologging that occurred when training models on
tf.data.Dataset
inputs (#6061, @dbczumar) - [Artifacts] Address artifact download failures from SFTP locations that occurred due to mismanaged concurrency (#5840, @rsundqvist)
- [Models] Fix a bug where MLflow Models did not restore bundled code properly if multiple models use the same code module name (#5926, @BFAnas)
- [Models] Address an issue where
mlflow.sklearn.model()
did not properly restore bundled model code (#6037, @WeichenXu123) - [Models] Fix a bug in
mlflow.evaluate()
that caused input data objects to be mutated when evaluating certain scikit-learn models (#6141, @dbczumar) - [Models] Fix a failure in
mlflow.pyfunc.spark_udf
that occurred when the UDF was invoked on an empty RDD partition (#6063, @WeichenXu123) - [Models] Fix a failure in
mlflow models build-docker
that occurred whenenv-manager=local
was specified (#6046, @bneijt) - [Projects] Improve robustness of the git repository check that occurs prior to MLflow Project execution (#6000, @dkapur17)
- [Projects] Address a failure that arose when running a Project that does not have a
master
branch (#5889, @harupy) - [Docs] Correct several typos throughout the MLflow docs (#5959, @ryanrussell)
Small bug fixes and doc updates (#6041, @drsantos89; #6138, #6137, #6132, @sunishsheth2009; #6144, #6124, #6125, #6123, #6057, #6060, #6050, #6038, #6029, #6030, #6025, #6018, #6019, #5962, #5974, #5972, #5957, #5947, #5907, #5938, #5906, #5932, #5919, #5914, #5888, #5890, #5886, #5873, #5865, #5843, @harupy; #6113, @comojin1994; #5930, @yashaswikakumanu; #5837, @shrinath-suresh; #6067, @deepyaman; #5997, @idlefella; #6021, @BenWilson2; #5984, @Sumanth077; #5929, @krunal16-c; #5879, @kugland; #5875, @ognis1205; #6006, @ryanrussell; #6140, @jinzhang21; #5983, @elk15; #6022, @apurva-koti; #5982, @EB-Joel; #5981, #5980, @punitkashyup; #6103, @ikrizanic; #5988, #5969, @SaumyaBhushan; #6020, #5991, @WeichenXu123; #5910, #5912, @Dark-Knight11; #6005, @Asinsa; #6023, @subramaniam02; #5999, @Regis-Caelum; #6007, @CaioCavalcanti; #5943, @kvaithin; #6017, #6002, @NeoKish; #6111, @T1b4lt; #5986, @seyyidibrahimgulec; #6053, @Zohair-coder; #6146, #6145, #6143, #6139, #6134, #6136, #6135, #6133, #6071, #6070, @dbczumar; #6026, @rotate2050)
MLflow 1.26.1 is a patch release containing the following bug fixes:
- [Installation] Fix compatibility issue with
protobuf >= 4.21.0
(#5945, @harupy) - [Models] Fix
get_model_dependencies
behavior formodels:
URIs containing artifact paths (#5921, @harupy) - [Models] Revert a problematic change to
artifacts
persistence inmlflow.pyfunc.log_model()
that was introduced in MLflow 1.25.0 (#5891, @kyle-jarvis) - [Models] Close associated image files when
EvaluationArtifact
outputs frommlflow.evaluate()
are garbage collected (#5900, @WeichenXu123)
Small bug fixes and updates (#5874, #5942, #5941, #5940, #5938, @harupy; #5893, @PrajwalBorkar; #5909, @yashaswikakumanu; #5937, @BenWilson2)
MLflow 1.26.0 includes several major features and improvements:
Features:
- [CLI] Add endpoint naming and options configuration to the deployment CLI (#5731, @trangevi)
- [Build,Doc] Add development environment setup script for Linux and MacOS x86 Operating Systems (#5717, @BenWilson2)
- [Tracking] Update
mlflow.set_tracking_uri
to add support for paths defined aspathlib.Path
in addition to existingstr
path declarations (#5824, @cacharle) - [Scoring] Add custom timeout override option to the scoring server CLI to support high latency models (#5663, @sniafas)
- [UI] Add sticky header to experiment run list table to support column name visibility when scrolling beyond page fold (#5818, @hubertzub-db)
- [Artifacts] Add GCS support for MLflow garbage collection (#5811, @aditya-iyengar-rtl-de)
- [Evaluate] Add
pos_label
argument foreval_and_log_metrics
API to support accurate binary classifier evaluation metrics (#5807, @yxiong) - [UI] Add fields for latest, minimum and maximum metric values on metric display page (#5574, @adamreeve)
- [Models] Add support for
input_example
andsignature
logging for pyspark ml flavor when using autologging (#5719, @bali0019) - [Models] Add
virtualenv
environment manager support formlflow models docker-build
CLI (#5728, @harupy) - [Models] Add support for wildcard module matching in log_model_allowlist for PySpark models (#5723, @serena-ruan)
- [Projects] Add
virtualenv
environment manager support for MLflow projects (#5631, @harupy) - [Models] Add
virtualenv
environment manager support for MLflow Models (#5380, @harupy) - [Models] Add
virtualenv
environment manager support formlflow.pyfunc.spark_udf
(#5676, @WeichenXu123) - [Models] Add support for
input_example
andsignature
logging fortensorflow
flavor when using autologging (#5510, @bali0019) - [Server-infra] Add JSON Schema Type Validation to enable raising 400 errors on malformed requests to REST API endpoints (#5458, @mrkaye97)
- [Scoring] Introduce abstract
endpoint
interface for mlflow deployments (#5378, @trangevi) - [UI] Add
End Time
andDuration
fields to run comparison page (#3378, @RealArpanBhattacharya) - [Serving] Add schema validation support when parsing input csv data for model serving (#5531, @vvijay-bolt)
Bug fixes and documentation updates:
- [Models] Fix REPL ID propagation from datasource listener to publisher for Spark data sources (#5826, @dbczumar)
- [UI] Update
ag-grid
and implementgetRowId
to improve performance in the runs table visualization (#5725, @adamreeve) - [Serving] Fix
tf-serving
parsing to support columnar-based formatting (#5825, @arjundc-db) - [Artifacts] Update
log_artifact
to support models larger than 2GB in HDFS (#5812, @hitchhicker) - [Models] Fix autologging to support
lightgbm
metric names with "@" symbols within their names (#5785, @mengchendd) - [Models] Pyfunc: Fix code directory resolution of subdirectories (#5806, @dbczumar)
- [Server-Infra] Fix mlflow-R server starting failure on windows (#5767, @serena-ruan)
- [Docs] Add documentation for
virtualenv
environment manager support for MLflow projects (#5727, @harupy) - [UI] Fix artifacts display sizing to support full width rendering in preview pane (#5606, @szczeles)
- [Models] Fix local hostname issues when loading spark model by binding driver address to localhost (#5753, @WeichenXu123)
- [Models] Fix autologging validation and batch_size calculations for
tensorflow
flavor (#5683, @MarkYHZhang) - [Artifacts] Fix
SqlAlchemyStore.log_batch
implementation to make it log data in batches (#5460, @erensahin)
Small bug fixes and doc updates (#5858, #5859, #5853, #5854, #5845, #5829, #5842, #5834, #5795, #5777, #5794, #5766, #5778, #5765, #5763, #5768, #5769, #5760, #5727, #5748, #5726, #5721, #5711, #5710, #5708, #5703, #5702, #5696, #5695, #5669, #5670, #5668, #5661, #5638, @harupy; #5749, @arpitjasa-db; #5675, @Davidswinkels; #5803, #5797, @ahlag; #5743, @kzhang01; #5650, #5805, #5724, #5720, #5662, @BenWilson2; #5627, @cterrelljones; #5646, @kutal10; #5758, @davideli-db; #5810, @rahulporuri; #5816, #5764, @shrinath-suresh; #5869, #5715, #5737, #5752, #5677, #5636, @WeichenXu123; #5735, @subramaniam02; #5746, @akaigraham; #5734, #5685, @lucalves; #5761, @marcelatoffernet; #5707, @aashish-khub; #5808, @ketangangal; #5730, #5700, @shaikmoeed; #5775, @dbczumar; #5747, @zhixuanevelynwu)
MLflow 1.25.1 is a patch release containing the following bug fixes:
- [Models] Fix a
pyfunc
artifact overwrite bug for when multiple artifacts are saved in sub-directories (#5657, @kyle-jarvis) - [Scoring] Fix permissions issue for Spark workers accessing model artifacts from a temp directory created by the driver (#5684, @WeichenXu123)
MLflow 1.25.0 includes several major features and improvements:
Features:
- [Tracking] Introduce a new fluent API
mlflow.last_active_run()
that provides the most recent fluent active run (#5584, @MarkYHZhang) - [Tracking] Add
experiment_names
argument to themlflow.search_runs()
API to support searching runs by experiment names (#5564, @r3stl355) - [Tracking] Add a
description
parameter tomlflow.start_run()
(#5534, @dogeplusplus) - [Tracking] Add
log_every_n_step
parameter tomlflow.pytorch.autolog()
to control metric logging frequency (#5516, @adamreeve) - [Tracking] Log
pyspark.ml.param.Params
values as MLflow parameters during PySpark autologging (#5481, @serena-ruan) - [Tracking] Add support for
pyspark.ml.Transformer
s to PySpark autologging (#5466, @serena-ruan) - [Tracking] Add input example and signature autologging for Keras models (#5461, @bali0019)
- [Models] Introduce
mlflow.diviner
flavor for large-scale time series forecasting (#5553, @BenWilson2) - [Models] Add
pyfunc.get_model_dependencies()
API to retrieve reproducible environment specifications for MLflow Models with the pyfunc flavor (#5503, @WeichenXu123) - [Models] Add
code_paths
argument to all model flavors to support packaging custom module code with MLflow Models (#5448, @stevenchen-db) - [Models] Support creating custom artifacts when evaluating models with
mlflow.evaluate()
(#5405, #5476 @MarkYHZhang) - [Models] Add
mlflow_version
field to MLModel specification (#5515, #5576, @r3stl355) - [Models] Add support for logging models to preexisting destination directories (#5572, @akshaya-a)
- [Scoring / Projects] Introduce
--env-manager
configuration for specifying environment restoration tools (e.g.conda
) and deprecate--no-conda
(#5567, @harupy) - [Scoring] Support restoring model dependencies in
mlflow.pyfunc.spark_udf()
to ensure accurate predictions (#5487, #5561, @WeichenXu123) - [Scoring] Add support for
numpy.ndarray
type inputs to the TensorFlow pyfuncpredict()
function (#5545, @WeichenXu123) - [Scoring] Support deployment of MLflow Models to Sagemaker Serverless (#5610, @matthewmayo)
- [UI] Add MLflow version to header beneath logo (#5504, @adamreeve)
- [Artifacts] Introduce a
mlflow.artifacts.download_artifacts()
API mirroring the functionality of themlflow artifacts download
CLI (#5585, @dbczumar) - [Artifacts] Introduce environment variables for controlling GCS artifact upload/download chunk size and timeouts (#5438, #5483, @mokrueger)
Bug fixes and documentation updates:
- [Tracking/SQLAlchemy] Create an index on
run_uuid
for PostgreSQL to improve query performance (#5446, @harupy) - [Tracking] Remove client-side validation of metric, param, tag, and experiment fields (#5593, @BenWilson2)
- [Projects] Support setting the name of the MLflow Run when executing an MLflow Project (#5187, @bramrodenburg)
- [Scoring] Use pandas
split
orientation for DataFrame inputs to SageMaker deploymentpredict()
API to preserve column ordering (#5522, @dbczumar) - [Server-Infra] Fix runs search compatibility bugs with PostgreSQL, MySQL, and MSSQL (#5540, @harupy)
- [CLI] Fix a bug in the
mlflow-skinny
client that causedmlflow --version
to fail (#5573, @BenWilson2) - [Docs] Update guidance and examples for model deployment to AzureML to recommend using the
mlflow-azureml
package (#5491, @santiagxf)
Small bug fixes and doc updates (#5591, #5629, #5597, #5592, #5562, #5477, @BenWilson2; #5554, @juntai-zheng; #5570, @tahesse; #5605, @guelate; #5633, #5632, #5625, #5623, #5615, #5608, #5600, #5603, #5602, #5596, #5587, #5586, #5580, #5577, #5568, #5290, #5556, #5560, #5557, #5548, #5547, #5538, #5513, #5505, #5464, #5495, #5488, #5485, #5468, #5455, #5453, #5454, #5452, #5445, #5431, @harupy; #5640, @nchittela; #5520, #5422, @Ark-kun; #5639, #5604, @nishipy; #5543, #5532, #5447, #5435, @WeichenXu123; #5502, @singankit; #5500, @Sohamkayal4103; #5449, #5442, @apurva-koti; #5552, @vinijaiswal; #5511, @adamreeve; #5428, @jinzhang21; #5309, @sunishsheth2009; #5581, #5559, @Kr4is; #5626, #5618, #5529, @sisp; #5652, #5624, #5622, #5613, #5509, #5459, #5437, @dbczumar; #5616, @liangz1)
MLflow 1.24.0 includes several major features and improvements:
Features:
- [Tracking] Support uploading, downloading, and listing artifacts through the MLflow server via
mlflow server --serve-artifacts
(#5320, @BenWilson2, @harupy) - [Tracking] Add the
registered_model_name
argument tomlflow.autolog()
for automatic model registration during autologging (#5395, @WeichenXu123) - [UI] Improve and restructure the Compare Runs page. Additions include "show diff only" toggles and scrollable tables (#5306, @WeichenXu123)
- [Models] Introduce
mlflow.pmdarima
flavor for pmdarima models (#5373, @BenWilson2) - [Models] When loading an MLflow Model, print a warning if a mismatch is detected between the current environment and the Model's dependencies (#5368, @WeichenXu123)
- [Models] Support computing custom scalar metrics during model evaluation with
mlflow.evaluate()
(#5389, @MarkYHZhang) - [Scoring] Add support for deploying and evaluating SageMaker models via the
MLflow Deployments API
(#4971, #5396, @jamestran201)
Bug fixes and documentation updates:
- [Tracking / UI] Fix artifact listing and download failures that occurred when operating the MLflow server in
--serve-artifacts
mode (#5409, @dbczumar) - [Tracking] Support environment-variable-based authentication when making artifact requests to the MLflow server in
--serve-artifacts
mode (#5370, @TimNooren) - [Tracking] Fix bugs in hostname and path resolution when making artifacts requests to the MLflow server in
--serve-artifacts
mode (#5384, #5385, @mert-kirpici) - [Tracking] Fix an import error that occurred when
mlflow.log_figure()
was used withoutmatplotlib.figure
imported (#5406, @WeichenXu123) - [Tracking] Correctly log XGBoost metrics containing the
@
symbol during autologging (#5403, @maxfriedrich) - [Tracking] Fix a SQL Server database error that occurred during Runs search (#5382, @dianacarvalho1)
- [Tracking] When downloading artifacts from HDFS, store them in the user-specified destination directory (#5210, @DimaClaudiu)
- [Tracking / Model Registry] Improve performance of large artifact and model downloads (#5359, @mehtayogita)
- [Models] Fix fast.ai PyFunc inference behavior for models with 2D outputs (#5411, @santiagxf)
- [Models] Record Spark model information to the active run when
mlflow.spark.log_model()
is called (#5355, @szczeles) - [Models] Restore onnxruntime execution providers when loading ONNX models with
mlflow.pyfunc.load_model()
(#5317, @ecm200) - [Projects] Increase Docker image push timeout when using Projects with Docker (#5363, @zanitete)
- [Python] Fix a bug that prevented users from enabling DEBUG-level Python log outputs (#5362, @dbczumar)
- [Docs] Add a developer guide explaining how to build custom plugins for
mlflow.evaluate()
(#5333, @WeichenXu123)
Small bug fixes and doc updates (#5298, @wamartin-aml; #5399, #5321, #5313, #5307, #5305, #5268, #5284, @harupy; #5329, @Ark-kun; #5375, #5346, #5304, @dbczumar; #5401, #5366, #5345, @BenWilson2; #5326, #5315, @WeichenXu123; #5236, @singankit; #5302, @timvink; #5357, @maitre-matt; #5347, #5344, @mehtayogita; #5367, @apurva-koti; #5348, #5328, #5310, @liangz1; #5267, @sunishsheth2009)
MLflow 1.23.1 is a patch release containing the following bug fixes:
- [Models] Fix a directory creation failure when loading PySpark ML models (#5299, @arjundc-db)
- [Model Registry] Revert to using case-insensitive validation logic for stage names in
models:/
URIs (#5312, @lichenran1234) - [Projects] Fix a race condition during Project tar file creation (#5303, @dbczumar)
MLflow 1.23.0 includes several major features and improvements:
Features:
- [Models] Introduce an
mlflow.evaluate()
API for evaluating MLflow Models, providing performance and explainability insights. For an overview, see https://mlflow.org/docs/latest/models.html#model-evaluation (#5069, #5092, #5256, @WeichenXu123) - [Models]
log_model()
APIs now return information about the logged MLflow Model, including artifact location, flavors, and schema (#5230, @liangz1) - [Models] Introduce an
mlflow.models.Model.load_input_example()
Python API for loading MLflow Model input examples (#5212, @maitre-matt) - [Models] Add a UUID field to the MLflow Model specification. MLflow Models now have a unique identifier (#5149, #5167, @WeichenXu123)
- [Models] Support passing SciPy CSC and CSR matrices as MLflow Model input examples (#5016, @WeichenXu123)
- [Model Registry] Support specifying
latest
in model URI to get the latest version of a model regardless of the stage (#5027, @lichenran1234) - [Tracking] Add support for LightGBM scikit-learn models to
mlflow.lightgbm.autolog()
(#5130, #5200, #5271 @jwyyy) - [Tracking] Improve S3 artifact download speed by caching boto clients (#4695, @Samreay)
- [UI] Automatically update metric plots for in-progress runs (#5017, @cedkoffeto, @harupy)
Bug fixes and documentation updates:
- [Models] Fix a bug in MLflow Model schema enforcement where strings were incorrectly cast to Pandas objects (#5134, @stevenchen-db)
- [Models] Fix a bug where keyword arguments passed to
mlflow.pytorch.load_model()
were not applied for scripted models (#5163, @schmidt-jake) - [Model Registry/R] Fix bug in R client
mlflow_create_model_version()
API that caused modelsource
to be set incorrectly (#5185, @bramrodenburg) - [Projects] Fix parsing behavior for Project URIs containing quotes (#5117, @dinaldoap)
- [Scoring] Use the correct 400-level error code for malformed MLflow Model Server requests (#5003, @abatomunkuev)
- [Tracking] Fix a bug where
mlflow.start_run()
modified user-supplied tags dictionary (#5191, @matheusMoreno) - [UI] Fix a bug causing redundant scroll bars to be displayed on the Experiment Page (#5159, @sunishsheth2009)
Small bug fixes and doc updates (#5275, #5264, #5244, #5249, #5255, #5248, #5243, #5240, #5239, #5232, #5234, #5235, #5082, #5220, #5219, #5226, #5217, #5194, #5188, #5132, #5182, #5183, #5180, #5177, #5165, #5164, #5162, #5015, #5136, #5065, #5125, #5106, #5127, #5120, @harupy; #5045, @BenWilson2; #5156, @pbezglasny; #5202, @jwyyy; #3863, @JoshuaAnickat; #5205, @abhiramr; #4604, @OSobky; #4256, @einsmein; #5140, @AveshCSingh; #5273, #5186, #5176, @WeichenXu123; #5260, #5229, #5206, #5174, #5160, @liangz1)
MLflow 1.22.0 includes several major features and improvements:
Features:
- [UI] Add a share button to the Experiment page (#4936, @marijncv)
- [UI] Improve readability of column sorting dropdown on Experiment page (#5022, @WeichenXu123; #5018, @NieuweNils, @coder-freestyle)
- [Tracking] Mark all autologging integrations as stable by removing
@experimental
decorators (#5028, @liangz1) - [Tracking] Add optional
experiment_id
parameter tomlflow.set_experiment()
(#5012, @dbczumar) - [Tracking] Add support for XGBoost scikit-learn models to
mlflow.xgboost.autolog()
(#5078, @jwyyy) - [Tracking] Improve statsmodels autologging performance by removing unnecessary metrics (#4942, @WeichenXu123)
- [Tracking] Update R client to tag nested runs with parent run ID (#4197, @yitao-li)
- [Models] Support saving and loading all XGBoost model types (#4954, @jwyyy)
- [Scoring] Support specifying AWS account and role when deploying models to SageMaker (#4923, @andresionek91)
- [Scoring] Support serving MLflow models with MLServer (#4963, @adriangonz)
Bug fixes and documentation updates:
- [UI] Fix bug causing Metric Plot page to crash when metric values are too large (#4947, @ianshan0915)
- [UI] Fix bug causing parallel coordinate curves to vanish (#5087, @harupy)
- [UI] Remove
Creator
field from Model Version page if user information is absent (#5089, @jinzhang21) - [UI] Fix model loading instructions for non-pyfunc models in Artifact Viewer (#5006, @harupy)
- [Models] Fix a bug that added
mlflow
toconda.yaml
even if a hashed version was already present (#5058, @maitre-matt) - [Docs] Add Python documentation for metric, parameter, and tag key / value length limits (#4991, @westford14)
- [Examples] Update Python version used in Prophet example to fix installation errors (#5101, @BenWilson2)
- [Examples] Fix Kubernetes
resources
specification in MLflow Projects + Kubernetes example (#4948, @jianyuan)
Small bug fixes and doc updates (#5119, #5107, #5105, #5103, #5085, #5088, #5051, #5081, #5039, #5073, #5072, #5066, #5064, #5063, #5060, #4718, #5053, #5052, #5041, #5043, #5047, #5036, #5037, #5029, #5031, #5032, #5030, #5007, #5019, #5014, #5008, #4998, #4985, #4984, #4970, #4966, #4980, #4967, #4978, #4979, #4968, #4976, #4975, #4934, #4956, #4938, #4950, #4946, #4939, #4913, #4940, #4935, @harupy; #5095, #5070, #5002, #4958, #4945, @BenWilson2; #5099, @chaosddp; #5005, @you-n-g; #5042, #4952, @shrinath-suresh; #4962, #4995, @WeichenXu123; #5010, @lichenran1234; #5000, @wentinghu; #5111, @alexott; #5102, #5024, #5011, #4959, @dbczumar; #5075, #5044, #5026, #4997, #4964, #4989, @liangz1; #4999, @stevenchen-db)
MLflow 1.21.0 includes several major features and improvements:
Features:
- [UI] Add a diff-only toggle to the runs table for filtering out columns with constant values (#4862, @marijncv)
- [UI] Add a duration column to the runs table (#4840, @marijncv)
- [UI] Display the default column sorting order in the runs table (#4847, @marijncv)
- [UI] Add
start_time
andduration
information to exported runs CSV (#4851, @marijncv) - [UI] Add lifecycle stage information to the run page (#4848, @marijncv)
- [UI] Collapse run page sections by default for space efficiency, limit artifact previews to 50MB (#4917, @dbczumar)
- [Tracking] Introduce autologging capabilities for PaddlePaddle model training (#4751, @jinminhao)
- [Tracking] Add an optional tags field to the CreateExperiment API (#4788, @dbczumar; #4795, @apurva-koti)
- [Tracking] Add support for deleting artifacts from SFTP stores via the
mlflow gc
CLI (#4670, @afaul) - [Tracking] Support AzureDefaultCredential for authenticating with Azure artifact storage backends (#4002, @marijncv)
- [Models] Upgrade the fastai model flavor to support fastai V2 (
>=2.4.1
) (#4715, @jinzhang21) - [Models] Introduce an
mlflow.prophet
model flavor for Prophet time series models (#4773, @BenWilson2) - [Models] Introduce a CLI for publishing MLflow Models to the SageMaker Model Registry (#4669, @jinnig)
- [Models] Print a warning when inferred model dependencies are not available on PyPI (#4891, @dbczumar)
- [Models, Projects] Add
MLFLOW_CONDA_CREATE_ENV_CMD
for customizing Conda environment creation (#4746, @giacomov)
Bug fixes and documentation updates:
- [UI] Fix an issue where column selections made in the runs table were persisted across experiments (#4926, @sunishsheth2009)
- [UI] Fix an issue where the text
null
was displayed in the runs table column ordering dropdown (#4924, @harupy) - [UI] Fix a bug causing the metric plot view to display NaN values upon click (#4858, @arpitjasa-db)
- [Tracking] Fix a model load failure for paths containing spaces or special characters on UNIX systems (#4890, @BenWilson2)
- [Tracking] Correct a migration issue that impacted usage of MLflow Tracking with SQL Server (#4880, @marijncv)
- [Tracking] Spark datasource autologging tags now respect the maximum allowable size for MLflow Tracking (#4809, @dbczumar)
- [Model Registry] Add previously-missing certificate sources for Model Registry REST API requests (#4731, @ericgosno91)
- [Model Registry] Throw an exception when users supply invalid Model Registry URIs for Databricks (#4877, @yunpark93)
- [Scoring] Fix a schema enforcement error that incorrectly cast date-like strings to datetime objects (#4902, @wentinghu)
- [Docs] Expand the documentation for the MLflow Skinny Client (#4113, @eedeleon)
Small bug fixes and doc updates (#4928, #4919, #4927, #4922, #4914, #4899, #4893, #4894, #4884, #4864, #4823, #4841, #4817, #4796, #4797, #4767, #4768, #4757, @harupy; #4863, #4838, @marijncv; #4834, @ksaur; #4772, @louisguitton; #4801, @twsl; #4929, #4887, #4856, #4843, #4789, #4780, @WeichenXu123; #4769, @Ark-kun; #4898, #4756, @apurva-koti; #4784, @lakshikaparihar; #4855, @ianshan0915; #4790, @eedeleon; #4931, #4857, #4846, 4777, #4748, @dbczumar)
MLflow 1.20.2 is a patch release containing the following features and bug fixes:
Features:
- Enabled auto dependency inference in spark flavor in autologging (#4759, @harupy)
Bug fixes and documentation updates:
- Increased MLflow client HTTP request timeout from 10s to 120s (#4764, @jinzhang21)
- Fixed autologging compatibility bugs with TensorFlow and Keras version
2.6.0
(#4766, @dbczumar)
Small bug fixes and doc updates (#4770, @WeichenXu123)
MLflow 1.20.1 is a patch release containing the following bug fixes:
- Avoid calling
importlib_metadata.packages_distributions
uponmlflow.utils.requirements_utils
import (#4741, @dbczumar) - Avoid depending on
importlib_metadata==4.7.0
(#4740, @dbczumar)
MLflow 1.20.0 includes several major features and improvements:
Features:
- Autologging for scikit-learn now records post training metrics when scikit-learn evaluation APIs, such as
sklearn.metrics.mean_squared_error
, are called (#4491, #4628 #4638, @WeichenXu123) - Autologging for PySpark ML now records post training metrics when model evaluation APIs, such as
Evaluator.evaluate()
, are called (#4686, @WeichenXu123) - Add
pip_requirements
andextra_pip_requirements
tomlflow.*.log_model
andmlflow.*.save_model
for directly specifying the pip requirements of the model to log / save (#4519, #4577, #4602, @harupy) - Added
stdMetrics
entries to the training metrics recorded during PySpark CrossValidator autologging (#4672, @WeichenXu123) - MLflow UI updates:
- Improved scalability of the parallel coordinates plot for run performance comparison,
- Added support for filtering runs based on their start time on the experiment page,
- Added a dropdown for runs table column sorting on the experiment page,
- Upgraded the AG Grid plugin, which is used for runs table loading on the experiment page, to version 25.0.0,
- Fixed a bug on the experiment page that caused the metrics section of the runs table to collapse when selecting columns from other table sections (#4712, @dbczumar)
- Added support for distributed execution to autologging for PyTorch Lightning (#4717, @dbczumar)
- Expanded R support for Model Registry functionality (#4527, @bramrodenburg)
- Added model scoring server support for defining custom prediction response wrappers (#4611, @Ark-kun)
mlflow.*.log_model
andmlflow.*.save_model
now automatically infer the pip requirements of the model to log / save based on the current software environment (#4518, @harupy)- Introduced support for running Sagemaker Batch Transform jobs with MLflow Models (#4410, #4589, @YQ-Wang)
Bug fixes and documentation updates:
- Deprecate
requirements_file
argument formlflow.*.save_model
andmlflow.*.log_model
(#4620, @harupy) - set nextPageToken to null (#4729, @harupy)
- Fix a bug in MLflow UI where the pagination token for run search is not refreshed when switching experiments (#4709, @harupy)
- Fix a bug in the model scoring server that rejected requests specifying a valid
Content-Type
header with the charset parameter (#4609, @Ark-kun) - Fixed a bug that caused SQLAlchemy backends to exhaust DB connections. (#4663, @arpitjasa-db)
- Improve docker build procedures to raise exceptions if docker builds fail (#4610, @Ark-kun)
- Disable autologging for scikit-learn
cross_val_*
APIs, which are incompatible with autologging (#4590, @WeichenXu123) - Deprecate MLflow Models support for fast.ai V1 (#4728, @dbczumar)
- Deprecate the old Azure ML deployment APIs
mlflow.azureml.cli.build_image
andmlflow.azureml.build_image
(#4646, @trangevi) - Deprecate MLflow Models support for TensorFlow < 2.0 and Keras < 2.3 (#4716, @harupy)
Small bug fixes and doc updates (#4730, #4722, #4725, #4723, #4703, #4710, #4679, #4694, #4707, #4708, #4706, #4705, #4625, #4701, #4700, #4662, #4699, #4682, #4691, #4684, #4683, #4675, #4666, #4648, #4653, #4651, #4641, #4649, #4627, #4637, #4632, #4634, #4621, #4619, #4622, #4460, #4608, #4605, #4599, #4600, #4581, #4583, #4565, #4575, #4564, #4580, #4572, #4570, #4574, #4576, #4568, #4559, #4537, #4542, @harupy; #4698, #4573, @Ark-kun; #4674, @kvmakes; #4555, @vagoston; #4644, @zhengjxu; #4690, #4588, @apurva-koti; #4545, #4631, #4734, @WeichenXu123; #4633, #4292, @shrinath-suresh; #4711, @jinzhang21; #4688, @murilommen; #4635, @ryan-duve; #4724, #4719, #4640, #4639, #4629, #4612, #4613, #4586, @dbczumar)
MLflow 1.19.0 includes several major features and improvements:
Features:
-
Add support for plotting per-class feature importance computed on linear boosters in XGBoost autologging (#4523, @dbczumar)
-
Add
mlflow_create_registered_model
andmlflow_delete_registered_model
for R to create/delete registered models. -
Add support for setting tags while resuming a run (#4497, @dbczumar)
-
MLflow UI updates (#4490, @sunishsheth2009)
- Add framework for internationalization support.
- Move metric columns before parameter and tag columns in the runs table.
- Change the display format of run start time to elapsed time (e.g. 3 minutes ago) from timestamp (e.g. 2021-07-14 14:02:10) in the runs table.
Bug fixes and documentation updates:
- Fix a bug causing MLflow UI to crash when sorting a column containing both
NaN
and empty values (#3409, @harupy)
Small bug fixes and doc updates (#4541, #4534, #4533, #4517, #4508, #4513, #4512, #4509, #4503, #4486, #4493, #4469, @harupy; #4458, @KasirajanA; #4501, @jimmyxu-db; #4521, #4515, @jerrylian-db; #4359, @shrinath-suresh; #4544, @WeichenXu123; #4549, @smurching; #4554, @derkomai; #4506, @tomasatdatabricks; #4551, #4516, #4494, @dbczumar; #4511, @keypointt)
MLflow 1.18.0 includes several major features and improvements:
Features:
- Autologging performance improvements for XGBoost, LightGBM, and scikit-learn (#4416, #4473, @dbczumar)
- Add new PaddlePaddle flavor to MLflow Models (#4406, #4439, @jinminhao)
- Introduce paginated ListExperiments API (#3881, @wamartin-aml)
- Include Runtime version for MLflow Models logged on Databricks (#4421, @stevenchen-db)
- MLflow Models now log dependencies in pip requirements.txt format, in addition to existing conda format (#4409, #4422, @stevenchen-db)
- Add support for limiting the number child runs created by autologging for scikit-learn hyperparameter search models (#4382, @mohamad-arabi)
- Improve artifact upload / download performance on Databricks (#4260, @dbczumar)
- Migrate all model dependencies from conda to "pip" section (#4393, @WeichenXu123)
Bug fixes and documentation updates:
- Fix an MLflow UI bug that caused git source URIs to be rendered improperly (#4403, @takabayashi)
- Fix a bug that prevented reloading of MLflow Models based on the TensorFlow SavedModel format (#4223) (#4319, @saschaschramm)
- Fix a bug in the behavior of
KubernetesSubmittedRun.get_status()
for Kubernetes MLflow Project runs (#3962) (#4159, @jcasse) - Fix a bug in TLS verification for MLflow artifact operations on S3 (#4047, @PeterSulcs)
- Fix a bug causing the MLflow server to crash after deletion of the default experiment (#4352, @asaf400)
- Fix a bug causing
mlflow models serve
to crash on Windows 10 (#4377, @simonvanbernem) - Fix a crash in runs search when ordering by metric values against the MSSQL backend store (#2551) (#4238, @naor2013)
- Fix an autologging incompatibility issue with TensorFlow 2.5 (#4371, @dbczumar)
- Fix a bug in the
disable_for_unsupported_versions
autologging argument that caused library versions to be incorrectly compared (#4303, @WeichenXu123)
Small bug fixes and doc updates (#4405, @mohamad-arabi; #4455, #4461, #4459, #4464, #4453, #4444, #4449, #4301, #4424, #4418, #4417, #3759, #4398, #4389, #4386, #4385, #4384, #4380, #4373, #4378, #4372, #4369, #4348, #4364, #4363, #4349, #4350, #4174, #4285, #4341, @harupy; #4446, @kHarshit; #4471, @AveshCSingh; #4435, #4440, #4368, #4360, @WeichenXu123; #4431, @apurva-koti; #4428, @stevenchen-db; #4467, #4402, #4261, @dbczumar)
MLflow 1.17.0 includes several major features and improvements:
Features:
- Add support for hyperparameter-tuning models to
mlflow.pyspark.ml.autolog()
(#4270, @WeichenXu123)
Bug fixes and documentation updates:
- Fix PyTorch Lightning callback definition for compatibility with PyTorch Lightning 1.3.0 (#4333, @dbczumar)
- Fix a bug in scikit-learn autologging that omitted artifacts for unsupervised models (#4325, @dbczumar)
- Support logging
datetime.date
objects as part of model input examples (#4313, @vperiyasamy) - Implement HTTP request retries in the MLflow Java client for 500-level responses (#4311, @dbczumar)
- Include a community code of conduct (#4310, @dennyglee)
Small bug fixes and doc updates (#4276, #4263, @WeichenXu123; #4289, #4302, #3599, #4287, #4284, #4265, #4266, #4275, #4268, @harupy; #4335, #4297, @dbczumar; #4324, #4320, @tleyden)
MLflow 1.16.0 includes several major features and improvements:
Features:
- Add
mlflow.pyspark.ml.autolog()
API for autologging ofpyspark.ml
estimators (#4228, @WeichenXu123) - Add
mlflow.catboost.log_model
,mlflow.catboost.save_model
,mlflow.catboost.load_model
APIs for CatBoost model persistence (#2417, @harupy) - Enable
mlflow.pyfunc.spark_udf
to use column names from model signature by default (#4236, @Loquats) - Add
datetime
data type for model signatures (#4241, @vperiyasamy) - Add
mlflow.sklearn.eval_and_log_metrics
API that computes and logs metrics for the given scikit-learn model and labeled dataset. (#4218, @alkispoly-db)
Bug fixes and documentation updates:
- Fix a database migration error for PostgreSQL (#4211, @dolfinus)
- Fix autologging silent mode bugs (#4231, @dbczumar)
Small bug fixes and doc updates (#4255, #4252, #4254, #4253, #4242, #4247, #4243, #4237, #4233, @harupy; #4225, @dmatrix; #4206, @mlflow-automation; #4207, @shrinath-suresh; #4264, @WeichenXu123; #3884, #3866, #3885, @ankan94; #4274, #4216, @dbczumar)
MLflow 1.15.0 includes several features, bug fixes and improvements. Notably, it includes a number of improvements to MLflow autologging:
Features:
- Add
silent=False
option to all autologging APIs, to allow suppressing MLflow warnings and logging statements during autologging setup and training (#4173, @dbczumar) - Add
disable_for_unsupported_versions=False
option to all autologging APIs, to disable autologging for versions of ML frameworks that have not been explicitly tested against the current version of the MLflow client (#4119, @WeichenXu123)
Bug fixes:
- Autologged runs are now terminated when execution is interrupted via SIGINT (#4200, @dbczumar)
- The R
mlflow_get_experiment
API now returns the same tag structure asmlflow_list_experiments
andmlflow_get_run
(#4017, @lorenzwalthert) - Fix bug where
mlflow.tensorflow.autolog
would previously mutate the user-specified callbacks list when fittingtf.keras
models (#4195, @dbczumar) - Fix bug where SQL-backed MLflow tracking server initialization failed when using the MLflow skinny client (#4161, @eedeleon)
- Model version creation (e.g. via
mlflow.register_model
) now fails if the model version status is not READY (#4114, @ankit-db)
Small bug fixes and doc updates (#4191, #4149, #4162, #4157, #4155, #4144, #4141, #4138, #4136, #4133, #3964, #4130, #4118, @harupy; #4152, @mlflow-automation; #4139, @WeichenXu123; #4193, @smurching; #4029, @architkulkarni; #4134, @xhochy; #4116, @wenleix; #4160, @wentinghu; #4203, #4184, #4167, @dbczumar)
MLflow 1.14.1 is a patch release containing the following bug fix:
- Fix issues in handling flexible numpy datatypes in TensorSpec (#4147, @arjundc-db)
MLflow 1.14.0 includes several major features and improvements:
- MLflow's model inference APIs (
mlflow.pyfunc.predict
), built-in model serving tools (mlflow models serve
), and model signatures now support tensor inputs. In particular, MLflow now provides built-in support for scoring PyTorch, TensorFlow, Keras, ONNX, and Gluon models with tensor inputs. For more information, see https://mlflow.org/docs/latest/models.html#deploy-mlflow-models (#3808, #3894, #4084, #4068 @wentinghu; #4041 @tomasatdatabricks, #4099, @arjundc-db) - Add new
mlflow.shap.log_explainer
,mlflow.shap.load_explainer
APIs for logging and loadingshap.Explainer
instances (#3989, @vivekchettiar) - The MLflow Python client is now available with a reduced dependency set via the
mlflow-skinny
PyPI package (#4049, @eedeleon) - Add new
RequestHeaderProvider
plugin interface for passing custom request headers with REST API requests made by the MLflow Python client (#4042, @jimmyxu-db) mlflow.keras.log_model
now saves models in the TensorFlow SavedModel format by default instead of the older Keras H5 format (#4043, @harupy)mlflow_log_model
now supports logging MLeap models in R (#3819, @yitao-li)- Add
mlflow.pytorch.log_state_dict
,mlflow.pytorch.load_state_dict
for logging and loading PyTorch state dicts (#3705, @shrinath-suresh) mlflow gc
can now garbage-collect artifacts stored in S3 (#3958, @sklingel)
Bug fixes and documentation updates:
- Enable autologging for TensorFlow estimators that extend
tensorflow.compat.v1.estimator.Estimator
(#4097, @mohamad-arabi) - Fix for universal autolog configs overriding integration-specific configs (#4093, @dbczumar)
- Allow
mlflow.models.infer_signature
to handle dataframes containingpandas.api.extensions.ExtensionDtype
(#4069, @caleboverman) - Fix bug where
mlflow_restore_run
doesn't propagate theclient
parameter tomlflow_get_run
(#4003, @yitao-li) - Fix bug where scoring on served model fails when request data contains a string that looks like URL and pandas version is later than 1.1.0 (#3921, @Secbone)
- Fix bug causing
mlflow_list_experiments
to fail listing experiments with tags (#3942, @lorenzwalthert) - Fix bug where metrics plots are computed from incorrect target values in scikit-learn autologging (#3993, @mtrencseni)
- Remove redundant / verbose Python event logging message in autologging (#3978, @dbczumar)
- Fix bug where
mlflow_load_model
doesn't load metadata associated to MLflow model flavor in R (#3872, @yitao-li) - Fix
mlflow.spark.log_model
,mlflow.spark.load_model
APIs on passthrough-enabled environments against ACL'd artifact locations (#3443, @smurching)
Small bug fixes and doc updates (#4102, #4101, #4096, #4091, #4067, #4059, #4016, #4054, #4052, #4051, #4038, #3992, #3990, #3981, #3949, #3948, #3937, #3834, #3906, #3774, #3916, #3907, #3938, #3929, #3900, #3902, #3899, #3901, #3891, #3889, @harupy; #4014, #4001, @dmatrix; #4028, #3957, @dbczumar; #3816, @lorenzwalthert; #3939, @pauldj54; #3740, @jkthompson; #4070, #3946, @jimmyxu-db; #3836, @t-henri; #3982, @neo-anderson; #3972, #3687, #3922, @eedeleon; #4044, @WeichenXu123; #4063, @yitao-li; #3976, @whiteh; #4110, @tomasatdatabricks; #4050, @apurva-koti; #4100, #4084, @wentinghu; #3947, @vperiyasamy; #4021, @trangevi; #3773, @ankan94; #4090, @jinzhang21; #3918, @danielfrg)
MLflow 1.13.1 is a patch release containing bug fixes and small changes:
- Fix bug causing Spark autologging to ignore configuration options specified by
mlflow.autolog()
(#3917, @dbczumar) - Fix bugs causing metrics to be dropped during TensorFlow autologging (#3913, #3914, @dbczumar)
- Fix incorrect value of optimizer name parameter in autologging PyTorch Lightning (#3901, @harupy)
- Fix model registry database
allow_null_for_run_id
migration failure affecting MySQL databases (#3836, @t-henri) - Fix failure in
transition_model_version_stage
when uncanonical stage name is passed (#3929, @harupy) - Fix an undefined variable error causing AzureML model deployment to fail (#3922, @eedeleon)
- Reclassify scikit-learn as a pip dependency in MLflow Model conda environments (#3896, @harupy)
- Fix experiment view crash and artifact view inconsistency caused by artifact URIs with redundant slashes (#3928, @dbczumar)
MLflow 1.13 includes several major features and improvements:
Features:
New fluent APIs for logging in-memory objects as artifacts:
- Add
mlflow.log_text
which logs text as an artifact (#3678, @harupy) - Add
mlflow.log_dict
which logs a dictionary as an artifact (#3685, @harupy) - Add
mlflow.log_figure
which logs a figure object as an artifact (#3707, @harupy) - Add
mlflow.log_image
which logs an image object as an artifact (#3728, @harupy)
UI updates / fixes (#3867, @smurching):
- Add model version link in compact experiment table view
- Add logged/registered model links in experiment runs page view
- Enhance artifact viewer for MLflow models
- Model registry UI settings are now persisted across browser sessions
- Add model version
description
field to model version table
Autologging enhancements:
- Improve robustness of autologging integrations to exceptions (#3682, #3815, dbczumar; #3860, @mohamad-arabi; #3854, #3855, #3861, @harupy)
- Add
disable
configuration option for autologging (#3682, #3815, dbczumar; #3838, @mohamad-arabi; #3854, #3855, #3861, @harupy) - Add
exclusive
configuration option for autologging (#3851, @apurva-koti; #3869, @dbczumar) - Add
log_models
configuration option for autologging (#3663, @mohamad-arabi) - Set tags on autologged runs for easy identification (and add tags to start_run) (#3847, @dbczumar)
More features and improvements:
- Allow Keras models to be saved with
SavedModel
format (#3552, @skylarbpayne) - Add support for
statsmodels
flavor (#3304, @olbapjose) - Add support for nested-run in mlflow R client (#3765, @yitao-li)
- Deploying a model using
mlflow.azureml.deploy
now integrates better with the AzureML tracking/registry. (#3419, @trangevi) - Update schema enforcement to handle integers with missing values (#3798, @tomasatdatabricks)
Bug fixes and documentation updates:
- When running an MLflow Project on Databricks, the version of MLflow installed on the Databricks cluster will now match the version used to run the Project (#3880, @FlorisHoogenboom)
- Fix bug where metrics are not logged for single-epoch
tf.keras
training sessions (#3853, @dbczumar) - Reject boolean types when logging MLflow metrics (#3822, @HCoban)
- Fix alignment of Keras /
tf.Keras
metric history entries wheninitial_epoch
is different from zero. (#3575, @garciparedes) - Fix bugs in autologging integrations for newer versions of TensorFlow and Keras (#3735, @dbczumar)
- Drop global
filterwwarnings
module at import time (#3621, @jogo) - Fix bug that caused preexisting Python loggers to be disabled when using MLflow with the SQLAlchemyStore (#3653, @arthury1n)
- Fix
h5py
library incompatibility for exported Keras models (#3667, @tomasatdatabricks)
Small changes, bug fixes and doc updates (#3887, #3882, #3845, #3833, #3830, #3828, #3826, #3825, #3800, #3809, #3807, #3786, #3794, #3731, #3776, #3760, #3771, #3754, #3750, #3749, #3747, #3736, #3701, #3699, #3698, #3658, #3675, @harupy; #3723, @mohamad-arabi; #3650, #3655, @shrinath-suresh; #3850, #3753, #3725, @dmatrix; ##3867, #3670, #3664, @smurching; #3681, @sueann; #3619, @andrewnitu; #3837, @javierluraschi; #3721, @szczeles; #3653, @arthury1n; #3883, #3874, #3870, #3877, #3878, #3815, #3859, #3844, #3703, @dbczumar; #3768, @wentinghu; #3784, @HCoban; #3643, #3649, @arjundc-db; #3864, @AveshCSingh, #3756, @yitao-li)
MLflow 1.12.1 is a patch release containing bug fixes and small changes:
- Fix
run_link
for cross-workspace model versions (#3681, @sueann) - Remove hard dependency on matplotlib for sklearn autologging (#3703, @dbczumar)
- Do not disable existing loggers when initializing alembic (#3653, @arthury1n)
MLflow 1.12.0 includes several major features and improvements, in particular a number of improvements to autologging and MLflow's Pytorch integrations:
Features:
Autologging:
- Add universal
mlflow.autolog
which enables autologging for all supported integrations (#3561, #3590, @andrewnitu) - Add
mlflow.pytorch.autolog
API for automatic logging of metrics, params, and models from Pytorch Lightning training (#3601, @shrinath-suresh, #3636, @karthik-77). This API is also enabled bymlflow.autolog
. - Scikit-learn, XGBoost, and LightGBM autologging now support logging model signatures and input examples (#3386, #3403, #3449, @andrewnitu)
mlflow.sklearn.autolog
now supports logging metrics (e.g. accuracy) and plots (e.g. confusion matrix heat map) (#3423, #3327, @willzhan-db, @harupy)
PyTorch:
mlflow.pytorch.log_model
,mlflow.pytorch.load_model
now support logging/loading TorchScript models (#3557, @shrinath-suresh)mlflow.pytorch.log_model
supports passingrequirements_file
&extra_files
arguments to log additional artifacts along with a model (#3436, @shrinath-suresh)
More features and improvements:
- Add
mlflow.shap.log_explanation
for logging model explanations generated by SHAP (#3513, @harupy) log_model
andcreate_model_version
now supports anawait_creation_for
argument (#3376, @andychow-db)- Put preview paths before non-preview paths for backwards compatibility (#3648, @sueann)
- Clean up model registry endpoint and client method definitions (#3610, @sueann)
- MLflow deployments plugin now supports 'predict' CLI command (#3597, @shrinath-suresh)
- Support H2O for R (#3416, @yitao-li)
- Add
MLFLOW_S3_IGNORE_TLS
environment variable to enable skipping TLS verification of S3 endpoint (#3345, @dolfinus)
Bug fixes and documentation updates:
- Ensure that results are synced across distributed processes if ddp enabled (no-op else) (#3651, @SeanNaren)
- Remove optimizer step override to ensure that all accelerator cases are covered by base module (#3635, @SeanNaren)
- Fix
AttributeError
in keras autologgging (#3611, @sephib) - Scikit-learn autologging: Exclude feature extraction / selection estimator (#3600, @dbczumar)
- Scikit-learn autologging: Fix behavior when a child and its parent are both patched (#3582, @dbczumar)
- Fix a bug where
lightgbm.Dataset(None)
fails after runningmlflow.lightgbm.autolog
(#3594, @harupy) - Fix a bug where
xgboost.DMatrix(None)
fails after runningmlflow.xgboost.autolog
(#3584, @harupy) - Pass
docker_args
in non-synchronous mlflow project runs (#3563, @alfozan) - Fix a bug of
FTPArtifactRepository.log_artifacts
withartifact_path
keyword argument (issue #3388) (#3391, @kzm4269) - Exclude preprocessing & imputation steps from scikit-learn autologging (#3491, @dbczumar)
- Fix duplicate stderr logging during artifact logging and project execution in the R client (#3145, @yitao-li)
- Don't call
atexit.register(_flush_queue)
in__main__
scope ofmlflow/tensorflow.py
(#3410, @harupy) - Fix for restarting terminated run not setting status correctly (#3329, @apurva-koti)
- Fix model version run_link URL for some Databricks regions (#3417, @sueann)
- Skip JSON validation when endpoint is not MLflow REST API (#3405, @harupy)
- Document
mlflow-torchserve
plugin (#3634, @karthik-77) - Add
mlflow-elasticsearchstore
to the doc (#3462, @AxelVivien25) - Add code snippets for fluent and MlflowClient APIs (#3385, #3437, #3489 #3573, @dmatrix)
- Document
mlflow-yarn
backend (#3373, @fhoering) - Fix a breakage in loading Tensorflow and Keras models (#3667, @tomasatdatabricks)
Small bug fixes and doc updates (#3607, #3616, #3534, #3598, #3542, #3568, #3349, #3554, #3544, #3541, #3533, #3535, #3516, #3512, #3497, #3522, #3521, #3492, #3502, #3434, #3422, #3394, #3387, #3294, #3324, #3654, @harupy; #3451, @jgc128; #3638, #3632, #3608, #3452, #3399, @shrinath-suresh; #3495, #3459, #3662, #3668, #3670 @smurching; #3488, @edgan8; #3639, @karthik-77; #3589, #3444, #3276, @lorenzwalthert; #3538, #3506, #3509, #3507, #3510, #3508, @rahulporuri; #3504, @sbrugman; #3486, #3466, @apurva-koti; #3477, @juntai-zheng; #3617, #3609, #3605, #3603, #3560, @dbczumar; #3411, @danielvdende; #3377, @willzhan-db; #3420, #3404, @andrewnitu; #3591, @mateiz; #3465, @abawchen; #3543, @emptalk; #3302, @bramrodenburg; #3468, @ghisvail; #3496, @extrospective; #3549, #3501, #3435, @yitao-li; #3243, @OlivierBondu; #3439, @andrewnitu; #3651, #3635 @SeanNaren, #3470, @ankit-db)
MLflow 1.11.0 includes several major features and improvements:
Features:
- New
mlflow.sklearn.autolog()
API for automatic logging of metrics, params, and models from scikit-learn model training (#3287, @harupy; #3323, #3358 @dbczumar) - Registered model & model version creation APIs now support specifying an initial
description
(#3271, @sueann) - The R
mlflow_log_model
andmlflow_load_model
APIs now support XGBoost models (#3085, @lorenzwalthert) - New
mlflow.list_run_infos
fluent API for listing run metadata (#3183, @trangevi) - Added section for visualizing and comparing model schemas to model version and model-version-comparison UIs (#3209, @zhidongqu-db)
- Enhanced support for using the model registry across Databricks workspaces: support for registering models to a Databricks workspace from outside the workspace (#3119, @sueann), tracking run-lineage of these models (#3128, #3164, @ankitmathur-db; #3187, @harupy), and calling
mlflow.<flavor>.load_model
against remote Databricks model registries (#3330, @sueann) - UI support for setting/deleting registered model and model version tags (#3187, @harupy)
- UI support for archiving existing staging/production versions of a model when transitioning a new model version to staging/production (#3134, @harupy)
Bug fixes and documentation updates:
- Fixed parsing of MLflow project parameter values containing'=' (#3347, @dbczumar)
- Fixed a bug preventing listing of WASBS artifacts on the latest version of Azure Blob Storage (12.4.0) (#3348, @dbczumar)
- Fixed a bug where artifact locations become malformed when using an SFTP file store in Windows (#3168, @harupy)
- Fixed bug where
list_artifacts
returned incorrect results on GCS, preventing e.g. loading SparkML models from GCS (#3242, @santosh1994) - Writing and reading artifacts via
MlflowClient
to a DBFS location in a Databricks tracking server specified through thetracking_uri
parameter during the initialization ofMlflowClient
now works properly (#3220, @sueann) - Fixed bug where
FTPArtifactRepository
returned artifact locations as absolute paths, rather than paths relative to the artifact repository root (#3210, @shaneing), and bug where callinglog_artifacts
against an FTP artifact location copied the logged directory itself into the FTP location, rather than the contents of the directory. - Fixed bug where Databricks project execution failed due to passing of GET request params as part of the request body rather than as query parameters (#2947, @cdemonchy-pro)
- Fix bug where artifact viewer did not correctly render PDFs in MLflow 1.10 (#3172, @ankitmathur-db)
- Fixed parsing of
order_by
arguments to MLflow search APIs when ordering by fields whose names contain spaces (#3118, @jdlesage) - Fixed bug where MLflow model schema enforcement raised exceptions when validating string columns using pandas >= 1.0 (#3130, @harupy)
- Fixed bug where
mlflow.spark.log_model
did not save model signature and input examples (#3151, @harupy) - Fixed bug in runs UI where tags table did not reflect deletion of tags. (#3135, @ParseDark)
- Added example illustrating the use of RAPIDS with MLflow (#3028, @drobison00)
Small bug fixes and doc updates (#3326, #3344, #3314, #3289, #3225, #3288, #3279, #3265, #3263, #3260, #3255, #3267, #3266, #3264, #3256, #3253, #3231, #3245, #3191, #3238, #3192, #3188, #3189, #3180, #3178, #3166, #3181, #3142, #3165, #2960, #3129, #3244, #3359 @harupy; #3236, #3141, @AveshCSingh; #3295, #3163, @arjundc-db; #3241, #3200, @zhidongqu-db; #3338, #3275, @sueann; #3020, @magnus-m; #3322, #3219, @dmatrix; #3341, #3179, #3355, #3360, #3363 @smurching; #3124, @jdlesage; #3232, #3146, @ankitmathur-db; #3140, @andreakress; #3174, #3133, @mlflow-automation; #3062, @cafeal; #3193, @tomasatdatabricks; 3115, @fhoering; #3328, @apurva-koti; #3046, @OlivierBondu; #3194, #3158, @dmatrix; #3250, @shivp950; #3259, @simonhessner; #3357 @dbczumar)
MLflow 1.10.0 includes several major features and improvements, in particular the release of several new model registry Python client APIs.
Features:
MlflowClient.transition_model_version_stage
now supports anarchive_existing_versions
argument for archiving existing staging or production model versions when transitioning a new model version to staging or production (#3095, @harupy)- Added
set_registry_uri
,get_registry_uri
APIs. Setting the model registry URI causes fluent APIs likemlflow.register_model
to communicate with the model registry at the specified URI (#3072, @sueann) - Added paginated
MlflowClient.search_registered_models
API (#2939, #3023, #3027 @ankitmathur-db; #2966, @mparkhe) - Added syntax highlighting when viewing text files (YAML etc) in the MLflow runs UI (#3041, @harupy)
- Added REST API and Python client support for setting and deleting tags on model versions and registered models,
via the
MlflowClient.create_registered_model
,MlflowClient.create_model_version
,MlflowClient.set_registered_model_tag
,MlflowClient.set_model_version_tag
,MlflowClient.delete_registered_model_tag
, andMlflowClient.delete_model_version_tag
APIs (#3094, @zhidongqu-db)
Bug fixes and documentation updates:
- Removed usage of deprecated
aws ecr get-login
command inmlflow.sagemaker
(#3036, @mrugeles) - Fixed bug where artifacts could not be viewed and downloaded from the artifact UI when using Azure Blob Storage (#3014, @Trollgeir)
- Databricks credentials are now propagated to the project subprocess when running MLflow projects within a notebook (#3035, @smurching)
- Added docs explaining how to fetching an MLflow model from the model registry (#3000, @andychow-db)
Small bug fixes and doc updates (#3112, #3102, #3089, #3103, #3096, #3090, #3049, #3080, #3070, #3078, #3083, #3051, #3050, #2875, #2982, #2949, #3121 @harupy; #3082, @ankitmathur-db; #3084, #3019, @smurching)
MLflow 1.9.1 is a patch release containing a number of bug-fixes and improvements:
Bug fixes and improvements:
- Fixes
AttributeError
when pickling an instance of the PythonMlflowClient
class (#2955, @Polyphenolx) - Fixes bug that prevented updating model-version descriptions in the model registry UI (#2969, @AnastasiaKol)
- Fixes bug where credentials were not properly propagated to artifact CLI commands when logging artifacts from Java to the DatabricksArtifactRepository (#3001, @dbczumar)
- Removes use of new Pandas API in new MLflow model-schema functionality, so that it can be used with older Pandas versions (#2988, @aarondav)
Small bug fixes and doc updates (#2998, @dbczumar; #2999, @arjundc-db)
MLflow 1.9.0 includes numerous major features and improvements, and a breaking change to experimental APIs:
Breaking Changes:
- The
new_name
argument toMlflowClient.update_registered_model
has been removed. CallMlflowClient.rename_registered_model
instead. (#2946, @mparkhe) - The
stage
argument toMlflowClient.update_model_version
has been removed. CallMlflowClient.transition_model_version_stage
instead. (#2946, @mparkhe)
Features (MLflow Models and Flavors)
log_model
andsave_model
APIs now support saving model signatures (the model's input and output schema) and example input along with the model itself (#2698, #2775, @tomasatdatabricks). Model signatures are used to reorder and validate input fields when scoring/serving models using the pyfunc flavor,mlflow models
CLI commands, ormlflow.pyfunc.spark_udf
(#2920, @tomasatdatabricks and @aarondav)- Introduce fastai model persistence and autologging APIs under
mlflow.fastai
(#2619, #2689 @antoniomdk) - Add pluggable
mlflow.deployments
API and CLI for deploying models to custom serving tools, e.g. RedisAI (#2327, @hhsecond) - Enables loading and scoring models whose conda environments include dependencies in conda-forge (#2797, @dbczumar)
- Add support for scoring ONNX-persisted models that return Python lists (#2742, @andychow-db)
Features (MLflow Projects)
- Add plugin interface for executing MLflow projects against custom backends (#2566, @jdlesage)
- Add ability to specify additional cluster-wide Python and Java libraries when executing MLflow projects remotely on Databricks (#2845, @pogil)
- Allow running MLflow projects against remote artifacts stored in any location with a corresponding ArtifactRepository implementation (Azure Blob Storage, GCS, etc) (#2774, @trangevi)
- Allow MLflow projects running on Kubernetes to specify a different tracking server to log to via the
KUBE_MLFLOW_TRACKING_URI
for passing a different tracking server to the kubernetes job (#2874, @catapulta)
Features (UI)
- Significant performance and scalability improvements to metric comparison and scatter plots in the UI (#2447, @mjlbach)
- The main MLflow experiment list UI now includes a link to the model registry UI (#2805, @zhidongqu-db),
- Enable viewing PDFs logged as artifacts from the runs UI (#2859, @ankmathur96)
- UI accessibility improvements: better color contrast (#2872, @Zangr), add child roles to DOM elements (#2871, @Zangr)
Features (Tracking Client and Server)
- Adds ability to pass client certs as part of REST API requests when using the tracking or model registry APIs. (#2843, @PhilipMay)
- New community plugin: support for storing artifacts in Aliyun (Alibaba Cloud) (#2917, @SeaOfOcean)
- Infer and set content type and encoding of objects when logging models and artifacts to S3 (#2881, @hajapy)
- Adds support for logging artifacts to HDFS Federation ViewFs (#2782, @fhoering)
- Add healthcheck endpoint to the MLflow server at
/health
(#2725, @crflynn) - Improves performance of default file-based tracking storage backend by using LibYAML (if installed) to read experiment and run metadata (#2707, @Higgcz)
Bug fixes and documentation updates:
- Several UI fixes: remove margins around icon buttons (#2827, @harupy),
fix alignment issues in metric view (#2811, @zhidongqu-db), add handling of
NaN
values in metrics plot (#2773, @dbczumar), truncate run ID in the run name when comparing multiple runs (#2508, @harupy) - Database engine URLs are no longer logged when running
mlflow db upgrade
(#2849, @hajapy) - Updates
log_artifact
,log_model
APIs to consistently use posix paths, rather than OS-dependent paths, when computing artifact subpaths. (#2784, @mikeoconnor0308) - Fix
ValueError
when scoringtf.keras
1.X models usingmlflow.pyfunc.predict
(#2762, @juntai-zheng) - Fixes conda environment activation bug when running MLflow projects on Windows (#2731, @MynherVanKoek)
mlflow.end_run
will now clear the active run even if the run cannot be marked as terminated (e.g. because it's been deleted), (#2693, @ahmed-shariff)- Add missing documentation for
mlflow.spacy
APIs (#2771, @harupy)
Small bug fixes and doc updates (#2919, @willzhan-db; #2940, #2942, #2941, #2943, #2927, #2929, #2926, #2914, #2928, #2913, #2852, #2876, #2808, #2810, #2442, #2780, #2758, #2732, #2734, #2431, #2733, #2716, @harupy; #2915, #2897, @jwgwalton; #2856, @jkthompson; #2962, @hhsecond; #2873, #2829, #2582, @dmatrix; #2908, #2865, #2880, #2866, #2833, #2785, #2723, @smurching; #2906, @dependabot[bot]; #2724, @aarondav; #2896, @ezeeetm; #2741, #2721, @mlflow-automation; #2864, @tallen94; #2726, @crflynn; #2710, #2951 @mparkhe; #2935, #2921, @ankitmathur-db; #2963, #2739, @dbczumar; #2853, @stat4jason; #2709, #2792, @juntai-zheng @juntai-zheng; #2749, @HiromuHota; #2957, #2911, #2718, @arjundc-db; #2885, @willzhan-db; #2803, #2761, @pogil; #2392, @jnmclarty; #2794, @Zethson; #2766, #2916 @shubham769)
MLflow 1.8.0 includes several major features and improvements:
Features:
- Added
mlflow.azureml.deploy
API for deploying MLflow models to AzureML (#2375 @csteegz, #2711, @akshaya-a) - Added support for case-sensitive LIKE and case-insensitive ILIKE queries (e.g.
'params.framework LIKE '%sklearn%'
) with the SearchRuns API & UI when running against a SQLite backend (#2217, @t-henri; #2708, @mparkhe) - Improved line smoothing in MLflow metrics UI using exponential moving averages (#2620, @Valentyn1997)
- Added
mlflow.spacy
module with support for logging and loading spaCy models (#2242, @arocketman) - Parameter values that differ across runs are highlighted in run comparison UI (#2565, @gabrielbretschner)
- Added ability to compare source runs associated with model versions from the registered model UI (#2537, @juntai-zheng)
- Added support for alphanumerical experiment IDs in the UI. (#2568, @jonas)
- Added support for passing arguments to
docker run
when running docker-based MLflow projects (#2608, @ksanjeevan) - Added Windows support for
mlflow sagemaker build-and-push-container
CLI & API (#2500, @AndreyBulezyuk) - Improved performance of reading experiment data from local filesystem when LibYAML is installed (#2707, @Higgcz)
- Added a healthcheck endpoint to the REST API server at
/health
that always returns a 200 response status code, to be used to verify health of the server (#2725, @crflynn) - MLflow metrics UI plots now scale to rendering thousands of points using scattergl (#2447, @mjlbach)
Bug fixes:
- Fixed CLI summary message in
mlflow azureml build_image
CLI (#2712, @dbczumar) - Updated
examples/flower_classifier/score_images_rest.py
with multiple bug fixes (#2647, @tfurmston) - Fixed pip not found error while packaging models via
mlflow models build-docker
(#2699, @HiromuHota) - Fixed bug in
mlflow.tensorflow.autolog
causing erroneous deletion of TensorBoard logging directory (#2670, @dbczumar) - Fixed a bug that truncated the description of the
mlflow gc
subcommand inmlflow --help
(#2679, @dbczumar) - Fixed bug where
mlflow models build-docker
was failing due to incorrect Miniconda download URL (#2685, @michaeltinsley) - Fixed a bug in S3 artifact logging functionality where
MLFLOW_S3_ENDPOINT_URL
was ignored (#2629, @poppash) - Fixed a bug where Sqlite in-memory was not working as a tracking backend store by modifying DB upgrade logic (#2667, @dbczumar)
- Fixed a bug to allow numerical parameters with values >= 1000 in R
mlflow::mlflow_run()
API (#2665, @lorenzwalthert) - Fixed a bug where AWS creds was not found in the Windows platform due path differences (#2634, @AndreyBulezyuk)
- Fixed a bug to add pip when necessary in
_mlflow_conda_env
(#2646, @tfurmston) - Fixed error code to be more meaningful if input to model version is incorrect (#2625, @andychow-db)
- Fixed multiple bugs in model registry (#2638, @aarondav)
- Fixed support for conda env dicts with
mlflow.pyfunc.log_model
(#2618, @dbczumar) - Fixed a bug where hiding the start time column in the UI would also hide run selection checkboxes (#2559, @harupy)
Documentation updates:
- Added links to source code to mlflow.org (#2627, @harupy)
- Documented fix for pandas-records payload (#2660, @SaiKiranBurle)
- Fixed documentation bug in TensorFlow
load_model
utility (#2666, @pogil) - Added the missing Model Registry description and link on the first page (#2536, @dmatrix)
- Added documentation for expected datatype for step argument in
log_metric
to match REST API (#2654, @mparkhe) - Added usage of the model registry to the
log_model
function insklearn_elasticnet_wine/train.py
example (#2609, @netanel246)
Small bug fixes and doc updates (#2594, @Trollgeir; #2703,#2709, @juntai-zheng; #2538, #2632, @keigohtr; #2656, #2553, @lorenzwalthert; #2622, @pingsutw; #2615, #2600, #2533, @mlflow-automation; #1391, @sueann; #2613, #2598, #2534, #2723, @smurching; #2652, #2710, @mparkhe; #2706, #2653, #2639, @tomasatdatabricks; #2611, @9dogs; #2700, #2705, @aarondav; #2675, #2540, @mengxr; #2686, @RensDimmendaal; #2694, #2695, #2532, @dbczumar; #2733, #2716, @harupy; #2726, @crflynn; #2582, #2687, @dmatrix)
MLflow 1.7.2 is a patch release containing a minor change:
- Pin alembic version to 1.4.1 or below to prevent pep517-related installation errors (#2612, @smurching)
MLflow 1.7.1 is a patch release containing bug fixes and small changes:
- Remove usage of Nonnull annotations and findbugs dependency in Java package (#2583, @mparkhe)
- Add version upper bound (<=1.3.13) to sqlalchemy dependency in Python package (#2587, @smurching)
Other bugfixes and doc updates (#2595, @mparkhe; #2567, @jdlesage)
MLflow 1.7.0 includes several major features and improvements, and some notable breaking changes:
MLflow support for Python 2 is now deprecated and will be dropped in a future release. At that point, existing Python 2 workflows that use MLflow will continue to work without modification, but Python 2 users will no longer get access to the latest MLflow features and bugfixes. We recommend that you upgrade to Python 3 - see https://docs.python.org/3/howto/pyporting.html for a migration guide.
Breaking changes to Model Registry REST APIs:
Model Registry REST APIs have been updated to be more consistent with the other MLflow APIs. With this release Model Registry APIs are intended to be stable until the next major version.
- Python and Java client APIs for Model Registry have been updated to use the new REST APIs. When using an MLflow client with a server using updated REST endpoints, you won't need to change any code but will need to upgrade to a new client version. The client APIs contain deprecated arguments, which for this release are backward compatible, but will be dropped in future releases. (#2457, @tomasatdatabricks; #2502, @mparkhe).
- The Model Registry UI has been updated to use the new REST APIs (#2476 @aarondav; #2507, @mparkhe)
Other Features:
- Ability to click through to individual runs from metrics plot (#2295, @harupy)
- Added
mlflow gc
CLI for permanent deletion of runs (#2265, @t-henri) - Metric plot state is now captured in page URLs for easier link sharing (#2393, #2408, #2498 @smurching; #2459, @harupy)
- Added experiment management to MLflow UI (create/rename/delete experiments) (#2348, @ggliem)
- Ability to search for experiments by name in the UI (#2324, @ggliem)
- MLflow UI page titles now reflect the content displayed on the page (#2420, @AveshCSingh)
- Added a new
LogModel
REST API endpoint for capturing model metadata, and call it from the Python and R clients (#2369, #2430, #2468 @tomasatdatabricks) - Java Client API to download model artifacts from Model Registry (#2308, @andychow-db)
Bug fixes and documentation updates:
- Updated Model Registry documentation page with code snippets and examples (#2493, @dmatrix; #2517, @harupy)
- Better error message for Model Registry, when using incompatible backend server (#2456, @aarondav)
- matplotlib is no longer required to use XGBoost and LightGBM autologging (#2423, @harupy)
- Fixed bug where matplotlib figures were not closed in XGBoost and LightGBM autologging (#2386, @harupy)
- Fixed parameter reading logic to support param values with newlines in FileStore (#2376, @dbczumar)
- Improve readability of run table column selector nodes (#2388, @dbczumar)
- Validate experiment name supplied to
UpdateExperiment
REST API endpoint (#2357, @ggliem) - Fixed broken MLflow DB README link in CLI docs (#2377, @dbczumar)
- Change copyright year across docs to 2020 (#2349, @ParseThis)
Small bug fixes and doc updates (#2378, #2449, #2402, #2397, #2391, #2387, #2523, #2527 @harupy; #2314, @juntai-zheng; #2404, @andychow-db; #2343, @pogil; #2366, #2370, #2364, #2356, @AveshCSingh; #2373, #2365, #2363, @smurching; #2358, @jcuquemelle; #2490, @RensDimmendaal; #2506, @dbczumar; #2234 @Zangr; #2359 @lbernickm; #2525, @mparkhe)
MLflow 1.6.0 includes several new features, including a better runs table interface, a utility for easier parameter tuning, and automatic logging from XGBoost, LightGBM, and Spark. It also implements a long-awaited fix allowing @ symbols in database URLs. A complete list is below:
Features:
- Adds a new runs table column view based on
ag-grid
which adds functionality for nested runs, serverside sorting, column reordering, highlighting, and more. (#2251, @Zangr) - Adds contour plot to the run comparsion page to better support parameter tuning (#2225, @harupy)
- If you use EarlyStopping with Keras autologging, MLflow now automatically captures the best model trained and the associated metrics (#2301, #2219, @juntai-zheng)
- Adds autologging functionality for LightGBM and XGBoost flavors to log feature importance, metrics per iteration, the trained model, and more. (#2275, #2238, @harupy)
- Adds an experimental mlflow.spark.autolog() API for automatic logging of Spark datasource information to the current active run. (#2220, @smurching)
- Optimizes the file store to load less data from disk for each operation (#2339, @jonas)
- Upgrades from ubuntu:16.04 to ubuntu:18.04 when building a Docker image with
mlflow models build-docker
(#2256, @andychow-db)
Bug fixes and documentation updates:
- Fixes bug when running server against database URLs with @ symbols (#2289, @hershaw)
- Fixes model Docker image build on Windows (#2257, @jahas)
- Documents the SQL Server plugin (#2320, @avflor)
- Adds a help file for the R package (#2259, @lorenzwalthert)
- Adds an example of using the Search API to find the best performing model (#2313, @AveshCSingh)
- Documents how to write and use MLflow plugins (#2270, @smurching)
Small bug fixes and doc updates (#2293, #2328, #2244, @harupy; #2269, #2332, #2306, #2307, #2292, #2267, #2191, #2231, @juntai-zheng; #2325, @shubham769; #2291, @sueann; #2315, #2249, #2288, #2278, #2253, #2181, @smurching; #2342, @tomasatdatabricks; #2245, @dependabot[bot]; #2338, @jcuquemelle; #2285, @avflor; #2340, @pogil; #2237, #2226, #2243, #2272, #2286, @dbczumar; #2281, @renaudhager; #2246, @avaucher; #2258, @lorenzwalthert; #2261, @smith-kyle; 2352, @dbczumar)
MLflow 1.5.0 includes several major features and improvements:
New Model Flavors and Flavor Updates:
- New support for a LightGBM flavor (#2136, @harupy)
- New support for a XGBoost flavor (#2124, @harupy)
- New support for a Gluon flavor and autologging (#1973, @cosmincatalin)
- Runs automatically created by
mlflow.tensorflow.autolog()
andmlflow.keras.autolog()
(#2088) are now automatically ended after training and/or exporting your model. See thedocs
for more details (#2094, @juntai-zheng)
More features and improvements:
- When using the
mlflow server
CLI command, you can now expose metrics on/metrics
for Prometheus via the optional --activate-parameter argument (#2097, @t-henri) - The
mlflow ui
CLI command now has a--host
/-h
option to specify user-input IPs to bind to (#2176, @gandroz) - MLflow now supports pulling Git submodules while using MLflow Projects (#2103, @badc0re)
- New
mlflow models prepare-env
command to do any preparation necessary to initialize an environment. This allows distinguishing configuration and user errors during predict/serve time (#2040, @aarondav) - TensorFlow.Keras and Keras parameters are now logged by
autolog()
(#2119, @juntai-zheng) - MLflow
log_params()
will recognize Spark ML params as keys and will now extract only the name attribute (#2064, @tomasatdatabricks) - Exposes
mlflow.tracking.is_tracking_uri_set()
(#2026, @fhoering) - The artifact image viewer now displays "Loading..." when it is loading an image (#1958, @harupy)
- The artifact image view now supports animated GIFs (#2070, @harupy)
- Adds ability to mount volumes and specify environment variables when using mlflow with docker (#1994, @nlml)
- Adds run context for detecting job information when using MLflow tracking APIs within Databricks Jobs. The following job types are supported: notebook jobs, Python Task jobs (#2205, @dbczumar)
- Performance improvement when searching for runs (#2030, #2059, @jcuquemelle; #2195, @rom1504)
Bug fixes and documentation updates:
- Fixed handling of empty directories in FS based artifact repositories (#1891, @tomasatdatabricks)
- Fixed
mlflow.keras.save_model()
usage with DBFS (#2216, @andychow-db) - Fixed several build issues for the Docker image (#2107, @jimthompson5802)
- Fixed
mlflow_list_artifacts()
(R package) (#2200, @lorenzwalthert) - Entrypoint commands of Kubernetes jobs are now shell-escaped (#2160, @zanitete)
- Fixed project run Conda path issue (#2147, @Zangr)
- Fixed spark model load from model repository (#2175, @tomasatdatabricks)
- Stripped "dev" suffix from PySpark versions (#2137, @dbczumar)
- Fixed note editor on the experiment page (#2054, @harupy)
- Fixed
models serve
,models predict
CLI commands against models:/ URIs (#2067, @smurching) - Don't unconditionally format values as metrics in generic HtmlTableView component (#2068, @smurching)
- Fixed remote execution from Windows using posixpath (#1996, @aestene)
- Add XGBoost and LightGBM examples (#2186, @harupy)
- Add note about active run instantiation side effect in fluent APIs (#2197, @andychow-db)
- The tutorial page has been refactored to be be a 'Tutorials and Examples' page (#2182, @juntai-zheng)
- Doc enhancements for XGBoost and LightGBM flavors (#2170, @harupy)
- Add doc for XGBoost flavor (#2167, @harupy)
- Updated
active_run()
docs to clarify it cannot be used accessing current run data (#2138, @juntai-zheng) - Document models:/ scheme for URI for load_model methods (#2128, @stbof)
- Added an example using Prophet via pyfunc (#2043, @dr3s)
- Added and updated some screenshots and explicit steps for the model registry (#2086, @stbof)
Small bug fixes and doc updates (#2142, #2121, #2105, #2069, #2083, #2061, #2022, #2036, #1972, #2034, #1998, #1959, @harupy; #2202, @t-henri; #2085, @stbof; #2098, @AdamBarnhard; #2180, #2109, #1977, #2039, #2062, @smurching; #2013, @aestene; #2146, @joelcthomas; #2161, #2120, #2100, #2095, #2088, #2076, #2057, @juntai-zheng; #2077, #2058, #2027, @sueann; #2149, @zanitete; #2204, #2188, @andychow-db; #2110, #2053, @jdlesage; #2003, #1953, #2004, @Djailla; #2074, @nlml; #2116, @Silas-Asamoah; #1104, @jimthompson5802; #2072, @cclauss; #2221, #2207, #2157, #2132, #2114, #2063, #2065, #2055, @dbczumar; #2033, @cthoyt; #2048, @philip-khor; #2002, @jspoorta; #2000, @christang; #2078, @dennyglee; #1986, @vguerra; #2020, @dependabot[bot])
MLflow 1.4.0 includes several major features:
-
Model Registry (Beta). Adds an experimental model registry feature, where you can manage, version, and keep lineage of your production models. (#1943, @mparkhe, @Zangr, @sueann, @dbczumar, @smurching, @gioa, @clemens-db, @pogil, @mateiz; #1988, #1989, #1995, #2021, @mparkhe; #1983, #1982, #1967, @dbczumar)
-
TensorFlow updates
- MLflow Keras model saving, loading, and logging has been updated to be compatible with TensorFlow 2.0. (#1927, @juntai-zheng)
- Autologging for
tf.estimator
andtf.keras
models has been updated to be compatible with TensorFlow 2.0. The same functionalities of autologging in TensorFlow 1.x are available in TensorFlow 2.0, namely when fittingtf.keras
models and when exporting savedtf.estimator
models. (#1910, @juntai-zheng) - Examples and READMEs for both TensorFlow 1.X and TensorFlow 2.0 have been added to
mlflow/examples/tensorflow
. (#1946, @juntai-zheng)
More features and improvements:
- [API] Add functions
get_run
,get_experiment
,get_experiment_by_name
to the fluent API (#1923, @fhoering) - [UI] Use Plotly as artifact image viewer, which allows zooming and panning (#1934, @harupy)
- [UI] Support deleting tags from the run details page (#1933, @harupy)
- [UI] Enable scrolling to zoom in metric and run comparison plots (#1929, @harupy)
- [Artifacts] Add support of viewfs URIs for HDFS federation for artifacts (#1947, @t-henri)
- [Models] Spark UDFs can now be called with struct input if the underlying spark implementation supports it. The data is passed as a pandas DataFrame with column names matching those in the struct. (#1882, @tomasatdatabricks)
- [Models] Spark models will now load faster from DFS by skipping unnecessary copies (#2008, @tomasatdatabricks)
Bug fixes and documentation updates:
- [Projects] Make detection of
MLproject
files case-insensitive (#1981, @smurching) - [UI] Fix a bug where viewing metrics containing forward-slashes in the name would break the MLflow UI (#1968, @smurching)
- [CLI]
models serve
command now works in Windows (#1949, @rboyes) - [Scoring] Fix a dependency installation bug in Java MLflow model scoring server (#1913, @smurching)
Small bug fixes and doc updates (#1932, #1935, @harupy; #1907, @marnixkoops; #1911, @HackyRoot; #1931, @jmcarp; #2007, @deniskovalenko; #1966, #1955, #1952, @Djailla; #1915, @sueann; #1978, #1894, @smurching; #1940, #1900, #1904, @mparkhe; #1914, @jerrygb; #1857, @mengxr; #2009, @dbczumar)
MLflow 1.3.0 includes several major features and improvements:
Features:
- The Python client now supports logging & loading models using TensorFlow 2.0 (#1872, @juntai-zheng)
- Significant performance improvements when fetching runs and experiments in MLflow servers that use SQL database-backed storage (#1767, #1878, #1805 @dbczumar)
- New
GetExperimentByName
REST API endpoint, used in the Python client to speed upset_experiment
andget_experiment_by_name
(#1775, @smurching) - New
mlflow.delete_run
,mlflow.delete_experiment
fluent APIs in the Python client(#1396, @MerelTheisenQB) - New CLI command (
mlflow experiments csv
) to export runs of an experiment into a CSV (#1705, @jdlesage) - Directories can now be logged as artifacts via
mlflow.log_artifact
in the Python fluent API (#1697, @apurva-koti) - HTML and geojson artifacts are now rendered in the run UI (#1838, @sim-san; #1803, @spadarian)
- Keras autologging support for
fit_generator
Keras API (#1757, @charnger) - MLflow models packaged as docker containers can be executed via Google Cloud Run (#1778, @ngallot)
- Artifact storage configurations are propagated to containers when executing docker-based MLflow projects locally (#1621, @nlaille)
- The Python, Java, R clients and UI now retry HTTP requests on 429 (Too Many Requests) errors (#1846, #1851, #1858, #1859 @tomasatdatabricks; #1847, @smurching)
Bug fixes and documentation updates:
- The R
mlflow_list_artifact
API no longer throws when listing artifacts for an empty run (#1862, @smurching) - Fixed a bug preventing running the MLflow server against an MS SQL database (#1758, @sifanLV)
- MLmodel files (artifacts) now correctly display in the run UI (#1819, @ankitmathur-db)
- The Python
mlflow.start_run
API now throws when resuming a run whose experiment ID differs from the active experiment ID set viamlflow.set_experiment
(#1820, @mcminnra). MlflowClient.log_metric
now logs metric timestamps with millisecond (as opposed to second) resolution (#1804, @ustcscgyer)- Fixed bugs when listing (#1800, @ahutterTA) and downloading (#1890, @jdlesage) artifacts stored in HDFS.
- Fixed a bug preventing Kubernetes Projects from pushing to private Docker repositories (#1788, @dbczumar)
- Fixed a bug preventing deploying Spark models to AzureML (#1769, @Ben-Epstein)
- Fixed experiment id resolution in projects (#1715, @drewmcdonald)
- Updated parallel coordinates plot to show all fields available in compared runs (#1753, @mateiz)
- Streamlined docs for getting started with hosted MLflow (#1834, #1785, #1860 @smurching)
Small bug fixes and doc updates (#1848, @pingsutw; #1868, @iver56; #1787, @apurvakoti; #1741, #1737, @apurva-koti; #1876, #1861, #1852, #1801, #1754, #1726, #1780, #1807 @smurching; #1859, #1858, #1851, @tomasatdatabricks; #1841, @ankitmathur-db; #1744, #1746, #1751, @mateiz; #1821, #1730, @dbczumar; #1727, cfmcgrady; #1716, @axsaucedo; #1714, @fhoering; #1405, @ancasarb; #1502, @jimthompson5802; #1720, jke-zq; #1871, @mehdi254; #1782, @stbof)
MLflow 1.2 includes the following major features and improvements:
- Experiments now have editable tags and descriptions (#1630, #1632, #1678, @ankitmathur-db)
- Search latency has been significantly reduced in the SQLAlchemyStore (#1660, @t-henri)
More features and improvements
- Backend stores now support run tag values up to 5000 characters in length. Some store implementations may support longer tag values (#1687, @ankitmathur-db)
- Gunicorn options can now be configured for the
mlflow models serve
CLI with theGUNICORN_CMD_ARGS
environment variable (#1557, @LarsDu) - Jsonnet artifacts can now be previewed in the UI (#1683, @ankitmathur-db)
- Adds an optional
python_version
argument tomlflow_install
for specifying the Python version (e.g. "3.5") to use within the conda environment created for installing the MLflow CLI. Ifpython_version
is unspecified,mlflow_install
defaults to using Python 3.6. (#1722, @smurching)
Bug fixes and documentation updates
- [Tracking] The Autologging feature is now more resilient to tracking errors (#1690, @apurva-koti)
- [Tracking] The
runs
field in in theGetExperiment.Response
proto has been deprecated & will be removed in MLflow 2.0. Please use theSearch Runs
API for fetching runs instead (#1647, @dbczumar) - [Projects] Fixed a bug that prevented docker-based MLflow Projects from logging artifacts to the
LocalArtifactRepository
(#1450, @nlaille) - [Projects] Running MLflow projects with the
--no-conda
flag in R no longer requires Anaconda to be installed (#1650, @spadarian) - [Models/Scoring] Fixed a bug that prevented Spark UDFs from being loaded on Databricks (#1658, @smurching)
- [UI] AJAX requests made by the MLflow Server Frontend now specify correct MIME-Types (#1679, @ynotzort)
- [UI] Previews now render correctly for artifacts with uppercase file extensions (e.g.,
.JSON
,.YAML
) (#1664, @ankitmathur-db) - [UI] Fixed a bug that caused search API errors to surface a Niagara Falls page (#1681, @dbczumar)
- [Installation] MLflow dependencies are now selected properly based on the target installation platform (#1643, @akshaya-a)
- [UI] Fixed a bug where the "load more" button in the experiment view did not appear on browsers in Windows (#1718, @Zangr)
Small bug fixes and doc updates (#1663, #1719, @dbczumar; #1693, @max-allen-db; #1695, #1659, @smurching; #1675, @jdlesage; #1699, @ankitmathur-db; #1696, @aarondav; #1710, #1700, #1656, @apurva-koti)
MLflow 1.1 includes several major features and improvements:
In MLflow Tracking:
- Experimental support for autologging from Tensorflow and Keras. Using
mlflow.tensorflow.autolog()
will enable automatic logging of metrics and optimizer parameters from TensorFlow to MLflow. The feature will work with TensorFlow versions1.12 <= v < 2.0
. (#1520, #1601, @apurva-koti) - Parallel coordinates plot in the MLflow compare run UI. Adds out of the box support for a parallel coordinates plot. The plot allows users to observe relationships between a n-dimensional set of parameters to metrics. It visualizes all runs as lines that are color-coded based on the value of a metric (e.g. accuracy), and shows what parameter values each run took on. (#1497, @Zangr)
- Pandas based search API. Adds the ability to return the results of a search as a pandas dataframe using the new
mlflow.search_runs
API. (#1483, #1548, @max-allen-db) - Java fluent API. Adds a new set of APIs to create and log to MLflow runs. This API contrasts with the existing low level
MlflowClient
API which simply wraps the REST APIs. The new fluent API allows you to create and log runs similar to how you would using the Python fluent API. (#1508, @andrewmchen) - Run tags improvements. Adds the ability to add and edit tags from the run view UI, delete tags from the API, and view tags in the experiment search view. (#1400, #1426, @Zangr; #1548, #1558, @ankitmathur-db)
- Search API improvements. Adds order by and pagination to the search API. Pagination allows you to read a large set of runs in small page sized chunks. This allows clients and backend implementations to handle an unbounded set of runs in a scalable manner. (#1444, @sueann; #1437, #1455, #1482, #1485, #1542, @aarondav; #1567, @max-allen-db; #1217, @mparkhe)
- Windows support for running the MLflow tracking server and UI. (#1080, @akshaya-a)
In MLflow Projects:
- Experimental support to run Docker based MLprojects in Kubernetes. Adds the first fully open source remote execution backend for MLflow projects. With this, you can leverage elastic compute resources managed by kubernetes for their ML training purposes. For example, you can run grid search over a set of hyperparameters by running several instances of an MLproject in parallel. (#1181, @marcusrehm, @tomasatdatabricks, @andrewmchen; #1566, @stbof, @dbczumar; #1574 @dbczumar)
More features and improvements
In MLflow Tracking:
- Paginated “load more” and backend sorting for experiment search view UI. This change allows the UI to scalably display the sorted runs from large experiments. (#1564, @Zangr)
- Search results are encoded in the URL. This allows you to share searches through their URL and to deep link to them. (#1416, @apurva-koti)
- Ability to serve MLflow UI behind
jupyter-server-proxy
or outside of the root path/
. Previous to MLflow 1.1, the UI could only be hosted on/
since the Javascript makes requests directly to/ajax-api/...
. With this patch, MLflow will make requests toajax-api/...
or a path relative to where the HTML is being served. (#1413, @xhochy)
In MLflow Models:
- Update
mlflow.spark.log_model()
to accept descendants of pyspark.Model (#1519, @ankitmathur-db) - Support for saving custom Keras models with
custom_objects
. This field is semantically equivalent to custom_objects parameter ofkeras.models.load_model()
function (#1525, @ankitmathur-db) - New more performant split orient based input format for pyfunc scoring server (#1479, @lennon310)
- Ability to specify gunicorn server options for pyfunc scoring server built with
mlflow models build-docker
. #1428, @lennon310)
Bug fixes and documentation updates
- [Tracking] Fix database migration for MySQL.
mlflow db upgrade
should now work for MySQL backends. (#1404, @sueann) - [Tracking] Make CLI
mlflow server
andmlflow ui
commands to work with SQLAlchemy URIs that specify a database driver. (#1411, @sueann) - [Tracking] Fix usability bugs related to FTP artifact repository. (#1398, @kafendt; #1421, @nlaille)
- [Tracking] Return appropriate HTTP status codes for MLflowException (#1434, @max-allen-db)
- [Tracking] Fix sorting by user ID in the experiment search view. (#1401, @andrewmchen)
- [Tracking] Allow calling log_metric with NaNs and infs. (#1573, @tomasatdatabricks)
- [Tracking] Fixes an infinite loop in downloading artifacts logged via dbfs and retrieved via S3. (#1605, @sueann)
- [Projects] Docker projects should preserve directory structure (#1436, @ahutterTA)
- [Projects] Fix conda activation for newer versions of conda. (#1576, @avinashraghuthu, @smurching)
- [Models] Allow you to log Tensorflow keras models from the
tf.keras
module. (#1546, @tomasatdatabricks)
Small bug fixes and doc updates (#1463, @mateiz; #1641, #1622, #1418, @sueann; #1607, #1568, #1536, #1478, #1406, #1408, @smurching; #1504, @LizaShak; #1490, @acroz; #1633, #1631, #1603, #1589, #1569, #1526, #1446, #1438, @apurva-koti; #1456, @Taur1ne; #1547, #1495, @aarondav; #1610, #1600, #1492, #1493, #1447, @tomasatdatabricks; #1430, @javierluraschi; #1424, @nathansuh; #1488, @henningsway; #1590, #1427, @Zangr; #1629, #1614, #1574, #1521, #1522, @dbczumar; #1577, #1514, @ankitmathur-db; #1588, #1566, @stbof; #1575, #1599, @max-allen-db; #1592, @abaveja313; #1606, @andrewmchen)
MLflow 1.0 includes many significant features and improvements. From this version, MLflow is no longer beta, and all APIs except those marked as experimental are intended to be stable until the next major version. As such, this release includes a number of breaking changes.
Major features, improvements, and breaking changes
-
Support for recording, querying, and visualizing metrics along a new “step” axis (x coordinate), providing increased flexibility for examining model performance relative to training progress. For example, you can now record performance metrics as a function of the number of training iterations or epochs. MLflow 1.0’s enhanced metrics UI enables you to visualize the change in a metric’s value as a function of its step, augmenting MLflow’s existing UI for plotting a metric’s value as a function of wall-clock time. (#1202, #1237, @dbczumar; #1132, #1142, #1143, @smurching; #1211, #1225, @Zangr; #1372, @stbof)
-
Search improvements. MLflow 1.0 includes additional support in both the API and UI for searching runs within a single experiment or a group of experiments. The search filter API supports a simplified version of the
SQL WHERE
clause. In addition to searching using run's metrics and params, the API has been enhanced to support a subset of run attributes as well as user and system tags. For details see Search syntax and examples for programmatically searching runs. (#1245, #1272, #1323, #1326, @mparkhe; #1052, @Zangr; #1363, @aarondav) -
Logging metrics in batches. MLflow 1.0 now has a
runs/log-batch
REST API endpoint for logging multiple metrics, params, and tags in a single API request. The endpoint useful for performant logging of multiple metrics at the end of a model training epoch (see example), or logging of many input model parameters at the start of training. You can call this batched-logging endpoint from Python (mlflow.log_metrics
,mlflow.log_params
,mlflow.set_tags
), R (mlflow_log_batch
), and Java (MlflowClient.logBatch
). (#1214, @dbczumar; see 0.9.1 and 0.9.0 for other changes) -
Windows support for MLflow Tracking. The Tracking portion of the MLflow client is now supported on Windows. (#1171, @eedeleon, @tomasatdatabricks)
-
HDFS support for artifacts. Hadoop artifact repository with Kerberos authorization support was added, so you can use HDFS to log and retrieve models and other artifacts. (#1011, @jaroslawk)
-
CLI command to build Docker images for serving. Added an
mlflow models build-docker
CLI command for building a Docker image capable of serving an MLflow model. The model is served at port 8080 within the container by default. Note that this API is experimental and does not guarantee that the arguments nor format of the Docker container will remain the same. (#1329, @smurching, @tomasatdatabricks) -
New
onnx
model flavor for saving, loading, and evaluating ONNX models with MLflow. ONNX flavor APIs are available in themlflow.onnx
module. (#1127, @avflor, @dbczumar; #1388, #1389, @dbczumar) -
Major breaking changes:
-
Some of the breaking changes involve database schema changes in the SQLAlchemy tracking store. If your database instance's schema is not up-to-date, MLflow will issue an error at the start-up of
mlflow server
ormlflow ui
. To migrate an existing database to the newest schema, you can use themlflow db upgrade
CLI command. (#1155, #1371, @smurching; #1360, @aarondav) -
[Installation] The MLflow Python package no longer depends on
scikit-learn
,mleap
, orboto3
. If you want to use thescikit-learn
support, theMLeap
support, ors3
artifact repository /sagemaker
support, you will have to install these respective dependencies explicitly. (#1223, @aarondav) -
[Artifacts] In the Models API, an artifact's location is now represented as a URI. See the documentation for the list of accepted URIs. (#1190, #1254, @dbczumar; #1174, @dbczumar, @sueann; #1206, @tomasatdatabricks; #1253, @stbof)
-
The affected methods are:
- Python:
<model-type>.load_model
,azureml.build_image
,sagemaker.deploy
,sagemaker.run_local
,pyfunc._load_model_env
,pyfunc.load_pyfunc
, andpyfunc.spark_udf
- R:
mlflow_load_model
,mlflow_rfunc_predict
,mlflow_rfunc_serve
- CLI:
mlflow models serve
,mlflow models predict
,mlflow sagemaker
,mlflow azureml
(with the new--model-uri
option)
- Python:
-
To allow referring to artifacts in the context of a run, MLflow introduces a new URI scheme of the form
runs:/<run_id>/relative/path/to/artifact
. (#1169, #1175, @sueann)
-
-
[CLI]
mlflow pyfunc
andmlflow rfunc
commands have been unified asmlflow models
(#1257, @tomasatdatabricks; #1321, @dbczumar) -
[CLI]
mlflow artifacts download
,mlflow artifacts download-from-uri
andmlflow download
commands have been consolidated intomlflow artifacts download
(#1233, @sueann) -
[Runs] Expose
RunData
fields (metrics
,params
,tags
) as dictionaries. Note that themlflow.entities.RunData
constructor still accepts lists ofmetric
/param
/tag
entities. (#1078, @smurching) -
[Runs] Rename
run_uuid
torun_id
in Python, Java, and REST API. Where necessary, MLflow will continue to acceptrun_uuid
until MLflow 1.1. (#1187, @aarondav)
-
Other breaking changes
CLI:
- The
--file-store
option is deprecated inmlflow server
andmlflow ui
commands. (#1196, @smurching) - The
--host
and--gunicorn-opts
options are removed in themlflow ui
command. (#1267, @aarondav) - Arguments to
mlflow experiments
subcommands, notably--experiment-name
and--experiment-id
are now options (#1235, @sueann) mlflow sagemaker list-flavors
has been removed (#1233, @sueann)
Tracking:
- The
user
property ofRun
s has been moved to tags (similarly, therun_name
,source_type
,source_name
properties were moved to tags in 0.9.0). (#1230, @acroz; #1275, #1276, @aarondav) - In R, the return values of experiment CRUD APIs have been updated to more closely match the REST API. In particular,
mlflow_create_experiment
now returns a string experiment ID instead of an experiment, and the other APIs return NULL. (#1246, @smurching) RunInfo.status
's type is now string. (#1264, @mparkhe)- Remove deprecated
RunInfo
properties fromstart_run
. (#1220, @aarondav) - As deprecated in 0.9.1 and before, the
RunInfo
fieldsrun_name
,source_name
,source_version
,source_type
, andentry_point_name
and theSearchRuns
fieldanded_expressions
have been removed from the REST API and Python, Java, and R tracking client APIs. They are still available as tags, documented in the REST API documentation. (#1188, @aarondav)
Models and deployment:
-
In Python, require arguments as keywords in
log_model
,save_model
andadd_to_model
methods in thetensorflow
andmleap
modules to avoid breaking changes in the future (#1226, @sueann) -
Remove the unsupported
jars
argument from ``spark.log_model` in Python (#1222, @sueann) -
Introduce
pyfunc.load_model
to be consistent with other Models modules.pyfunc.load_pyfunc
will be deprecated in the near future. (#1222, @sueann) -
Rename
dst_path
parameter inpyfunc.save_model
topath
(#1221, @aarondav) -
R flavors refactor (#1299, @kevinykuo)
mlflow_predict()
has been added in favor ofmlflow_predict_model()
andmlflow_predict_flavor()
which have been removed.mlflow_save_model()
is now a generic andmlflow_save_flavor()
is no longer needed and has been removed.mlflow_predict()
takes...
to pass to underlying predict methods.mlflow_load_flavor()
now has the signaturefunction(flavor, model_path)
and flavor authors should implementmlflow_load_flavor.mlflow_flavor_{FLAVORNAME}
. The flavor argument is inferred from the inputs of user-facingmlflow_load_model()
and does not need to be explicitly provided by the user.
Projects:
- Remove and rename some
projects.run
parameters for generality and consistency. (#1222, @sueann) - In R, the
mlflow_run
API for running MLflow projects has been modified to more closely reflect the Pythonmlflow.run
API. In particular, the order of theuri
andentry_point
arguments has been reversed and theparam_list
argument has been renamed toparameters
. (#1265, @smurching)
R:
- Remove
mlflow_snapshot
andmlflow_restore_snapshot
APIs. Also, ther_dependencies
argument used to specify the path to a packrat r-dependencies.txt file has been removed from all APIs. (#1263, @smurching) - The
mlflow_cli
andcrate
APIs are now private. (#1246, @smurching)
Environment variables:
-
Prefix environment variables with "MLFLOW_" (#1268, @aarondav). Affected variables are:
- [Tracking]
_MLFLOW_SERVER_FILE_STORE
,_MLFLOW_SERVER_ARTIFACT_ROOT
,_MLFLOW_STATIC_PREFIX
- [SageMaker]
MLFLOW_SAGEMAKER_DEPLOY_IMG_URL
,MLFLOW_DEPLOYMENT_FLAVOR_NAME
- [Scoring]
MLFLOW_SCORING_SERVER_MIN_THREADS
,MLFLOW_SCORING_SERVER_MAX_THREADS
- [Tracking]
More features and improvements
- [Tracking] Non-default driver support for SQLAlchemy backends:
db+driver
is now a valid tracking backend URI scheme (#1297, @drewmcdonald; #1374, @mparkhe) - [Tracking] Validate backend store URI before starting tracking server (#1218, @luke-zhu, @sueann)
- [Tracking] Add
GetMetricHistory
client API in Python and Java corresponding to the REST API. (#1178, @smurching) - [Tracking] Add
view_type
argument toMlflowClient.list_experiments()
in Python. (#1212, @smurching) - [Tracking] Dictionary values provided to
mlflow.log_params
andmlflow.set_tags
in Python can now be non-string types (e.g., numbers), and they are automatically converted to strings. (#1364, @aarondav) - [Tracking] R API additions to be at parity with REST API and Python (#1122, @kevinykuo)
- [Tracking] Limit number of results returned from
SearchRuns
API and UI for faster load (#1125, @mparkhe; #1154, @andrewmchen) - [Artifacts] To avoid having many copies of large model files in serving,
ArtifactRepository.download_artifacts
no longer copies local artifacts (#1307, @andrewmchen; #1383, @dbczumar) - [Artifacts/Projects] Support GCS in download utilities.
gs://bucket/path
files are now supported by themlflow artifacts download
CLI command and as parameters of typepath
in MLProject files. (#1168, @drewmcdonald) - [Models] All Python models exported by MLflow now declare
mlflow
as a dependency by default. In addition, we introduce a flag--install-mlflow
users can pass tomlflow models serve
andmlflow models predict
methods to force installation of the latest version of MLflow into the model's environment. (#1308, @tomasatdatabricks) - [Models] Update model flavors to lazily import dependencies in Python. Modules that define Model flavors now import extra dependencies such as
tensorflow
,scikit-learn
, andpytorch
inside individual methods, ensuring that these modules can be imported and explored even if the dependencies have not been installed on your system. Also, theDEFAULT_CONDA_ENVIRONMENT
module variable has been replaced with aget_default_conda_env()
function for each flavor. (#1238, @dbczumar) - [Models] It is now possible to pass extra arguments to
mlflow.keras.load_model
that will be passed through tokeras.load_model
. (#1330, @yorickvP) - [Serving] For better performance, switch to
gunicorn
for serving Python models. This does not change the user interface. (#1322, @tomasatdatabricks) - [Deployment] For SageMaker, use the uniquely-generated model name as the S3 bucket prefix instead of requiring one. (#1183, @dbczumar)
- [REST API] Add support for API paths without the
preview
component. Thepreview
paths will be deprecated in a future version of MLflow. (#1236, @mparkhe)
Bug fixes and documentation updates
- [Tracking] Log metric timestamps in milliseconds by default (#1177, @smurching; #1333, @dbczumar)
- [Tracking] Fix bug when deserializing integer experiment ID for runs in
SQLAlchemyStore
(#1167, @smurching) - [Tracking] Ensure unique constraint names in MLflow tracking database (#1292, @smurching)
- [Tracking] Fix base64 encoding for basic auth in R tracking client (#1126, @freefrag)
- [Tracking] Correctly handle
file:
URIs for the-—backend-store-uri
option inmlflow server
andmlflow ui
CLI commands (#1171, @eedeleon, @tomasatdatabricks) - [Artifacts] Update artifact repository download methods to return absolute paths (#1179, @dbczumar)
- [Artifacts] Make FileStore respect the default artifact location (#1332, @dbczumar)
- [Artifacts] Fix
log_artifact
failures due to existing directory on FTP server (#1327, @kafendt) - [Artifacts] Fix GCS artifact logging of subdirectories (#1285, @jason-huling)
- [Projects] Fix bug not sharing
SQLite
database file with Docker container (#1347, @tomasatdatabricks; #1375, @aarondav) - [Java] Mark
sendPost
andsendGet
as experimental (#1186, @aarondav) - [Python/CLI] Mark
azureml.build_image
as experimental (#1222, #1233 @sueann) - [Docs] Document public MLflow environment variables (#1343, @aarondav)
- [Docs] Document MLflow system tags for runs (#1342, @aarondav)
- [Docs] Autogenerate CLI documentation to include subcommands and descriptions (#1231, @sueann)
- [Docs] Update run selection description in
mlflow_get_run
in R documentation (#1258, @dbczumar) - [Examples] Update examples to reflect API changes (#1361, @tomasatdatabricks; #1367, @mparkhe)
Small bug fixes and doc updates (#1359, #1350, #1331, #1301, #1270, #1271, #1180, #1144, #1135, #1131, #1358, #1369, #1368, #1387, @aarondav; #1373, @akarloff; #1287, #1344, #1309, @stbof; #1312, @hchiuzhuo; #1348, #1349, #1294, #1227, #1384, @tomasatdatabricks; #1345, @withsmilo; #1316, @ancasarb; #1313, #1310, #1305, #1289, #1256, #1124, #1097, #1162, #1163, #1137, #1351, @smurching; #1319, #1244, #1224, #1195, #1194, #1328, @dbczumar; #1213, #1200, @Kublai-Jing; #1304, #1320, @andrewmchen; #1311, @Zangr; #1306, #1293, #1147, @mateiz; #1303, @gliptak; #1261, #1192, @eedeleon; #1273, #1259, @kevinykuo; #1277, #1247, #1243, #1182, #1376, @mparkhe; #1210, @vgod-dbx; #1199, @ashtuchkin; #1176, #1138, #1365, @sueann; #1157, @cclauss; #1156, @clemens-db; #1152, @pogil; #1146, @srowen; #875, #1251, @jimthompson5802)
MLflow 0.9.1 is a patch release on top of 0.9.0 containing mostly bug fixes and internal improvements. We have also included a one breaking API change in preparation for additions in MLflow 1.0 and later. This release also includes significant improvements to the Search API.
Breaking changes:
- [Tracking] Generalized experiment_id to string (from a long) to be more permissive of different ID types in different backend stores. While breaking for the REST API, this change is backwards compatible for python and R clients. (#1067, #1034 @eedeleon)
More features and improvements:
- [Search/API] Moving search filters into a query string based syntax, with Java client, Python client, and UI support. This also improves quote, period, and special character handling in query strings and adds the ability to search on tags in filter string. (#1042, #1055, #1063, #1068, #1099, #1106 @mparkhe; #1025 @andrewmchen; #1060 @smurching)
- [Tracking] Limits and validations to batch-logging APIs in OSS server (#958 @smurching)
- [Tracking/Java] Java client API for batch-logging (#1081 @mparkhe)
- [Tracking] Improved consistency of handling multiple metric values per timestamp across tracking stores (#972, #999 @dbczumar)
Bug fixes and documentation updates:
- [Tracking/Python] Reintroduces the parent_run_id argument to MlflowClient.create_run. This API is planned for removal in MLflow 1.0 (#1137 @smurching)
- [Tracking/Python] Provide default implementations of AbstractStore log methods (#1051 @acroz)
- [R] (Released on CRAN as MLflow 0.9.0.1) Small bug fixes with R (#1123 @smurching; #1045, #1017, #1019, #1039, #1048, #1098, #1101, #1107, #1108, #1119 @tomasatdatabricks)
Small bug fixes and doc updates (#1024, #1029 @bayethiernodiop; #1075 @avflor; #968, #1010, #1070, #1091, #1092 @smurching; #1004, #1085 @dbczumar; #1033, #1046 @sueann; #1053 @tomasatdatabricks; #987 @hanyucui; #935, #941 @jimthompson5802; #963 @amilbourne; #1016 @andrewmchen; #991 @jaroslawk; #1007 @mparkhe)
Bugfix release (PyPI only) with the following changes:
-
Rebuilt MLflow JS assets to fix an issue where form input was broken in MLflow 0.9.0 (identified in #1056, #1113 by @shu-yusa, @timothyjlaurent)
0.9.0 (2019-03-13)
Major features:
-
Support for running MLflow Projects in Docker containers. This allows you to include non-Python dependencies in their project environments and provides stronger isolation when running projects. See the Projects documentation for more information. (#555, @marcusrehm; #819, @mparkhe; #970, @dbczumar)
-
Database stores for the MLflow Tracking Server. Support for a scalable and performant backend store was one of the top community requests. This feature enables you to connect to local or remote SQLAlchemy-compatible databases (currently supported flavors include MySQL, PostgreSQL, SQLite, and MS SQL) and is compatible with file backed store. See the Tracking Store documentation for more information. (#756, @AndersonReyes; #800, #844, #847, #848, #860, #868, #975, @mparkhe; #980, @dbczumar)
-
Simplified custom Python model packaging. You can easily include custom preprocessing and postprocessing logic, as well as data dependencies in models with the
python_function
flavor using updatedmlflow.pyfunc
Python APIs. For more information, see the Custom Python Models documentation. (#791, #792, #793, #830, #910, @dbczumar) -
Plugin systems allowing third party libraries to extend MLflow functionality. The proposal document gives the full detail of the three main changes:
- You can register additional providers of tracking stores using the
mlflow.tracking_store
entrypoint. (#881, @zblz) - You can register additional providers of artifact repositories using the
mlflow.artifact_repository
entrypoint. (#882, @mociarain) - The logic generating run metadata from the run context (e.g.
source_name
,source_version
) has been refactored into an extendable system of run context providers. Plugins can register additional providers using themlflow.run_context_provider
entrypoint, which add to or overwrite tags set by the base library. (#913, #926, #930, #978, @acroz)
- You can register additional providers of tracking stores using the
-
Support for HTTP authentication to the Tracking Server in the R client. Now you can connect to secure Tracking Servers using credentials set in environment variables, or provide custom plugins for setting the credentials. As an example, this release contains a Databricks plugin that can detect existing Databricks credentials to allow you to connect to the Databricks Tracking Server. (#938, #959, #992, @tomasatdatabricks)
Breaking changes:
- [Scoring] The
pyfunc
scoring server now expects requests with theapplication/json
content type to contain json-serialized pandas dataframes in the split format, rather than the records format. See the documentation on deployment for more detail. (#960, @dbczumar) Also, when reading the pandas dataframes from JSON, the scoring server no longer automatically infers data types as it can result in unintentional conversion of data types (#916, @mparkhe). - [API] Remove
GetMetric
&GetParam
from the REST API as they are subsumed byGetRun
. (#879, @aarondav)
More features and improvements:
- [UI] Add a button for downloading artifacts (#967, @mateiz)
- [CLI] Add CLI commands for runs: now you can
list
,delete
,restore
, anddescribe
runs through the CLI (#720, @DorIndivo) - [CLI] The
run
command now can take--experiment-name
as an argument, as an alternative to the--experiment-id
argument. You can also choose to set the_EXPERIMENT_NAME_ENV_VAR
environment variable instead of passing in the value explicitly. (#889, #894, @mparkhe) - [Examples] Add Image classification example with Keras. (#743, @tomasatdatabricks )
- [Artifacts] Add
get_artifact_uri()
and_download_artifact_from_uri
convenience functions (#779) - [Artifacts] Allow writing Spark models directly to the target artifact store when possible (#808, @smurching)
- [Models] PyTorch model persistence improvements to allow persisting definitions and dependencies outside the immediate scope:
- Add a
code_paths
parameter tomlflow.pytorch.save_model
andmlflow.pytorch.log_model
to allow external module dependencies to be specified as paths to python files. (#842, @dbczumar) - Improve
mlflow.pytorch.save_model
to capture class definitions from notebooks and the__main__
scope (#851, #861, @dbczumar)
- Add a
- [Runs/R] Allow client to infer context info when creating new run in fluent API (#958, @tomasatdatabricks)
- [Runs/UI] Support Git Commit hyperlink for Gitlab and Bitbucket. Previously the clickable hyperlink was generated only for Github pages. (#901)
- [Search]/API] Allow param value to have any content, not just alphanumeric characters,
.
, and-
(#788, @mparkhe) - [Search/API] Support "filter" string in the
SearchRuns
API. Corresponding UI improvements are planned for the future (#905, @mparkhe) - [Logging] Basic support for LogBatch. NOTE: The feature is currently experimental and the behavior is expected to change in the near future. (#950, #951, #955, #1001, @smurching)
Bug fixes and documentation updates:
- [Artifacts] Fix empty-file upload to DBFS in
log_artifact
andlog_artifacts
(#895, #818, @smurching) - [Artifacts] S3 artifact store: fix path resolution error when artifact root is bucket root (#928, @dbczumar)
- [UI] Fix a bug with Databricks notebook URL links (#891, @smurching)
- [Export] Fix for missing run name in csv export (#864, @jimthompson5802)
- [Example] Correct missing tensorboardX module error in PyTorch example when running in MLflow Docker container (#809, @jimthompson5802)
- [Scoring/R] Fix local serving of rfunc models (#874, @kevinykuo)
- [Docs] Improve flavor-specific documentation in Models documentation (#909, @dbczumar)
Small bug fixes and doc updates (#822, #899, #787, #785, #780, #942, @hanyucui; #862, #904, #954, #806, #857, #845, @stbof; #907, #872, @smurching; #896, #858, #836, #859, #923, #939, #933, #931, #952, @dbczumar; #880, @zblz; #876, @acroz; #827, #812, #816, #829, @jimthompson5802; #837, #790, #897, #974, #900, @mparkhe; #831, #798, @aarondav; #814, @sueann; #824, #912, @mateiz; #922, #947, @tomasatdatabricks; #795, @KevYuen; #676, @mlaradji; #906, @4n4nd; #777, @tmielika; #804, @alkersan)
MLflow 0.8.2 is a patch release on top of 0.8.1 containing only bug fixes and no breaking changes or features.
Bug fixes:
- [Python API] CloudPickle has been added to the set of MLflow library dependencies, fixing missing import errors when attempting to save models (#777, @tmielika)
- [Python API] Fixed a malformed logging call that prevented
mlflow.sagemaker.push_image_to_ecr()
invocations from succeeding (#784, @jackblandin) - [Models] PyTorch models can now be saved with code dependencies, allowing model classes to be loaded successfully in new environments (#842, #836, @dbczumar)
- [Artifacts] Fixed a timeout when logging zero-length files to DBFS artifact stores (#818, @smurching)
Small docs updates (#845, @stbof; #840, @grahamhealy20; #839, @wilderrodrigues)
MLflow 0.8.1 introduces several significant improvements:
- Improved UI responsiveness and load time, especially when displaying experiments containing hundreds to thousands of runs.
- Improved visualizations, including interactive scatter plots for MLflow run comparisons
- Expanded support for scoring Python models as Spark UDFs. For more information, see the updated documentation for this feature.
- By default, saved models will now include a Conda environment specifying all of the dependencies necessary for loading them in a new environment.
Features:
- [API/CLI] Support for running MLflow projects from ZIP files (#759, @jmorefieldexpe)
- [Python API] Support for passing model conda environments as dictionaries to
save_model
andlog_model
functions (#748, @dbczumar) - [Models] Default Anaconda environments have been added to many Python model flavors. By default, models produced by
save_model
andlog_model
functions will include an environment that specifies all of the versioned dependencies necessary to load and serve the models. Previously, users had to specify these environments manually. (#705, #707, #708, #749, @dbczumar) - [Scoring] Support for synchronous deployment of models to SageMaker (#717, @dbczumar)
- [Tracking] Include the Git repository URL as a tag when tracking an MLflow run within a Git repository (#741, @whiletruelearn, @mateiz)
- [UI] Improved runs UI performance by using a react-virtualized table to optimize row rendering (#765, #762, #745, @smurching)
- [UI] Significant performance improvements for rendering run metrics, tags, and parameter information (#764, #747, @smurching)
- [UI] Scatter plots, including run comparsion plots, are now interactive (#737, @mateiz)
- [UI] Extended CSRF support by allowing the MLflow UI server to specify a set of expected headers that clients should set when making AJAX requests (#733, @aarondav)
Bug fixes and documentation updates:
- [Python/Scoring] MLflow Python models that produce Pandas DataFrames can now be evaluated as Spark UDFs correctly. Spark UDF outputs containing multiple columns of primitive types are now supported (#719, @tomasatdatabricks)
- [Scoring] Fixed a serialization error that prevented models served with Azure ML from returning Pandas DataFrames (#754, @dbczumar)
- [Docs] New example demonstrating how the MLflow REST API can be used to create experiments and log run information (#750, kjahan)
- [Docs] R documentation has been updated for clarity and style consistency (#683, @stbof)
- [Docs] Added clarification about user setup requirements for executing remote MLflow runs on Databricks (#736, @andyk)
Small bug fixes and doc updates (#768, #715, @smurching; #728, dodysw; #730, mshr-h; #725, @kryptec; #769, #721, @dbczumar; #714, @stbof)
MLflow 0.8.0 introduces several major features:
-
Dramatically improved UI for comparing experiment run results:
- Metrics and parameters are by default grouped into a single column, to avoid an explosion of mostly-empty columns. Individual metrics and parameters can be moved into their own column to help compare across rows.
- Runs that are "nested" inside other runs (e.g., as part of a hyperparameter search or multistep workflow) now show up grouped by their parent run, and can be expanded or collapsed altogether. Runs can be nested by calling
mlflow.start_run
ormlflow.run
while already within a run. - Run names (as opposed to automatically generated run UUIDs) now show up instead of the run ID, making comparing runs in graphs easier.
- The state of the run results table, including filters, sorting, and expanded rows, is persisted in browser local storage, making it easier to go back and forth between an individual run view and the table.
-
Support for deploying models as Docker containers directly to Azure Machine Learning Service Workspace (as opposed to the previously-recommended solution of Azure ML Workbench).
Breaking changes:
- [CLI]
mlflow sklearn serve
has been removed in favor ofmlflow pyfunc serve
, which takes the same arguments but works against any pyfunc model (#690, @dbczumar)
Features:
- [Scoring] pyfunc server and SageMaker now support the pandas "split" JSON format in addition to the "records" format. The split format allows the client to specify the order of columns, which is necessary for some model formats. We recommend switching client code over to use this new format (by sending the Content-Type header
application/json; format=pandas-split
), as it will become the default JSON format in MLflow 0.9.0. (#690, @dbczumar) - [UI] Add compact experiment view (#546, #620, #662, #665, @smurching)
- [UI] Add support for viewing & tracking nested runs in experiment view (#588, @andrewmchen; #618, #619, @aarondav)
- [UI] Persist experiments view filters and sorting in browser local storage (#687, @smurching)
- [UI] Show run name instead of run ID when present (#476, @smurching)
- [Scoring] Support for deploying Models directly to Azure Machine Learning Service Workspace (#631, @dbczumar)
- [Server/Python/Java] Add
rename_experiment
to Tracking API (#570, @aarondav) - [Server] Add
get_experiment_by_name
to RestStore (#592, @dmarkhas) - [Server] Allow passing gunicorn options when starting mlflow server (#626, @mparkhe)
- [Python] Cloudpickle support for sklearn serialization (#653, @dbczumar)
- [Artifacts] FTP artifactory store added (#287, @Shenggan)
Bug fixes and documentation updates:
- [Python] Update TensorFlow integration to match API provided by other flavors (#612, @dbczumar; #670, @mlaradji)
- [Python] Support for TensorFlow 1.12 (#692, @smurching)
- [R] Explicitly loading Keras module at predict time no longer required (#586, @kevinykuo)
- [R] pyfunc serve can correctly load models saved with the R Keras support (#634, @tomasatdatabricks)
- [R] Increase network timeout of calls to the RestStore from 1 second to 60 seconds (#704, @aarondav)
- [Server] Improve errors returned by RestStore (#582, @andrewmchen; #560, @smurching)
- [Server] Deleting the default experiment no longer causes it to be immediately recreated (#604, @andrewmchen; #641, @schipiga)
- [Server] Azure Blob Storage artifact repo supports Windows paths (#642, @marcusrehm)
- [Server] Improve behavior when environment and run files are corrupted (#632, #654, #661, @mparkhe)
- [UI] Improve error page when viewing nonexistent runs or views (#600, @andrewmchen; #560, @andrewmchen)
- [UI] UI no longer throws an error if all experiments are deleted (#605, @andrewmchen)
- [Docs] Include diagram of workflow for multistep example (#581, @dennyglee)
- [Docs] Add reference tags and R and Java APIs to tracking documentation (#514, @stbof)
- [Docs/R] Use CRAN installation (#686, @javierluraschi)
Small bug fixes and doc updates (#576, #594, @javierluraschi; #585, @kevinykuo; #593, #601, #611, #650, #669, #671, #679, @dbczumar; #607, @suzil; #583, #615, @andrewmchen; #622, #681, @aarondav; #625, @pogil; #589, @tomasatdatabricks; #529, #635, #684, @stbof; #657, @mvsusp; #682, @mateiz; #678, vfdev-5; #596, @yutannihilation; #663, @smurching)
MLflow 0.7.0 introduces several major features:
- An R client API (to be released on CRAN soon)
- Support for deleting runs (API + UI)
- UI support for adding notes to a run
The release also includes bugfixes and improvements across the Python and Java clients, tracking UI, and documentation.
Breaking changes:
- [Python] The per-flavor implementation of load_pyfunc has been made private (#539, @tomasatdatabricks)
- [REST API, Java] logMetric now accepts a double metric value instead of a float (#566, @aarondav)
Features:
- [R] Support for R (#370, #471, @javierluraschi; #548 @kevinykuo)
- [UI] Add support for adding notes to Runs (#396, @aadamson)
- [Python] Python API, REST API, and UI support for deleting Runs (#418, #473, #526, #579 @andrewmchen)
- [Python] Set a tag containing the branch name when executing a branch of a Git project (#469, @adrian555)
- [Python] Add a set_experiment API to activate an experiment before starting runs (#462, @mparkhe)
- [Python] Add arguments for specifying a parent run to tracking & projects APIs (#547, @andrewmchen)
- [Java] Add Java set tag API (#495, @smurching)
- [Python] Support logging a conda environment with sklearn models (#489, @dbczumar)
- [Scoring] Support downloading MLflow scoring JAR from Maven during scoring container build (#507, @dbczumar)
Bug fixes:
- [Python] Print errors when the Databricks run fails to start (#412, @andrewmchen)
- [Python] Fix Spark ML PyFunc loader to work on Spark driver (#480, @tomasatdatabricks)
- [Python] Fix Spark ML load_pyfunc on distributed clusters (#490, @tomasatdatabricks)
- [Python] Fix error when downloading artifacts from a run's artifact root (#472, @dbczumar)
- [Python] Fix DBFS upload file-existence-checking logic during Databricks project execution (#510, @smurching)
- [Python] Support multi-line and unicode tags (#502, @mparkhe)
- [Python] Add missing DeleteExperiment, RestoreExperiment implementations in the Python REST API client (#551, @mparkhe)
- [Scoring] Convert Spark DataFrame schema to an MLeap schema prior to serialization (#540, @dbczumar)
- [UI] Fix bar chart always showing in metric view (#488, @smurching)
Small bug fixes and doc updates (#467 @drorata; #470, #497, #508, #518 @dbczumar; #455, #466, #492, #504, #527 @aarondav; #481, #475, #484, #496, #515, #517, #498, #521, #522, #573 @smurching; #477 @parkerzf; #494 @jainr; #501, #531, #532, #552 @mparkhe; #503, #520 @dmatrix; #509, #532 @tomasatdatabricks; #484, #486 @stbof; #533, #534 @javierluraschi; #542 @GCBallesteros; #511 @AdamBarnhard)
MLflow 0.6.0 introduces several major features:
- A Java client API, available on Maven
- Support for saving and serving SparkML models as MLeap for low-latency serving
- Support for tagging runs with metadata, during and after the run completion
- Support for deleting (and restoring deleted) experiments
In addition to these features, there are a host of improvements and bugfixes to the REST API, Python API, tracking UI, and documentation. The examples subdirectory has also been revamped to make it easier to jump in, and examples demonstrating multistep workflows and hyperparameter tuning have been added.
Breaking changes:
We fixed a few inconsistencies in the the mlflow.tracking
API, as introduced in 0.5.0:
MLflowService
has been renamedMlflowClient
(#461, @mparkhe)- You get an
MlflowClient
by callingmlflow.tracking.MlflowClient()
(previously, this wasmlflow.tracking.get_service()
) (#461, @mparkhe) MlflowService.list_runs
was changed toMlflowService.list_run_infos
to reflect the information actually returned by the call. It now returns aRunInfo
instead of aRun
(#334, @aarondav)MlflowService.log_artifact
andMlflowService.log_artifacts
now take arun_id
instead ofartifact_uri
. This now matcheslist_artifacts
anddownload_artifacts
(#444, @aarondav)
Features:
- Java client API added with support for the MLflow Tracking API (analogous to
mlflow.tracking
), allowing users to create and manage experiments, runs, and artifacts. The release includes a usage exampleand Javadocs. The client is published to Maven undermlflow:mlflow
(#380, #394, #398, #409, #410, #430, #452, @aarondav) - SparkML models are now also saved in MLeap format (https://github.com/combust/mleap), when applicable. Model serving platforms can choose to serve using this format instead of the SparkML format to dramatically decrease prediction latency. SageMaker now does this by default (#324, #327, #331, #395, #428, #435, #438, @dbczumar)
- [API] Experiments can now be deleted and restored via REST API, Python Tracking API, and MLflow CLI (#340, #344, #367, @mparkhe)
- [API] Tags can now be set via a SetTag API, and they have been moved to
RunData
fromRunInfo
(#342, @aarondav) - [API] Added
list_artifacts
anddownload_artifacts
toMlflowService
to interact with a run's artifactory (#350, @andrewmchen) - [API] Added
get_experiment_by_name
to Python Tracking API, and equivalent to Java API (#373, @vfdev-5) - [API/Python] Version is now exposed via
mlflow.__version__
. - [API/CLI] Added
mlflow artifacts
CLI to list, download, and upload to run artifact repositories (#391, @aarondav) - [UI] Added icons to source names in MLflow Experiments UI (#381, @andrewmchen)
- [UI] Added support to view
.log
and.tsv
files from MLflow artifacts UI (#393, @Shenggan; #433, @whiletruelearn) - [UI] Run names can now be edited from within the MLflow UI (#382, @smurching)
- [Serving] Added
--host
option tomlflow serve
to allow listening on non-local addressess (#401, @hamroune) - [Serving/SageMaker] SageMaker serving takes an AWS region argument (#366, @dbczumar)
- [Python] Added environment variables to support providing HTTP auth (username, password, token) when talking to a remote MLflow tracking server (#402, @aarondav)
- [Python] Added support to override S3 endpoint for S3 artifactory (#451, @hamroune)
- MLflow nightly Python wheel and JAR snapshots are now available and linked from https://github.com/mlflow/mlflow (#352, @aarondav)
Bug fixes and documentation updates:
- [Python]
mlflow run
now logs default parameters, in addition to explicitly provided ones (#392, @mparkhe) - [Python]
log_artifact
in FileStore now requires a relative path as the artifact path (#439, @mparkhe) - [Python] Fixed string representation of Python entities, so they now display both their type and serialized fields (#371, @smurching)
- [UI] Entry point name is now shown in MLflow UI (#345, @aarondav)
- [Models] Keras model export now includes TensorFlow graph explicitly to ensure the model can always be loaded at deployment time (#440, @tomasatdatabricks)
- [Python] Fixed issue where FileStore ignored provided Run Name (#358, @adrian555)
- [Python] Fixed an issue where any
mlflow run
failing printed an extraneous exception (#365, @smurching) - [Python] uuid dependency removed (#351, @antonpaquin)
- [Python] Fixed issues with remote execution on Databricks (#357, #361, @smurching; #383, #387, @aarondav)
- [Docs] Added comprehensive example of doing a multistep workflow, chaining MLflow runs together and reusing results (#338, @aarondav)
- [Docs] Added comprehensive example of doing hyperparameter tuning (#368, @tomasatdatabricks)
- [Docs] Added code examples to
mlflow.keras
API (#341, @dmatrix) - [Docs] Significant improvements to Python API documentation (#454, @stbof)
- [Docs] Examples folder refactored to improve readability. The examples now reside in
examples/
instead ofexample/
, too (#399, @mparkhe) - Small bug fixes and doc updates (#328, #363, @ToonKBC; #336, #411, @aarondav; #284, @smurching; #377, @mparkhe; #389, gioa; #408, @aadamson; #397, @vfdev-5; #420, @adrian555; #459, #463, @stbof)
MLflow 0.5.2 is a patch release on top of 0.5.1 containing only bug fixes and no breaking changes or features.
Bug fixes:
- Fix a bug with ECR client creation that caused
mlflow.sagemaker.deploy()
to fail when searching for a deployment Docker image (#366, @dbczumar)
MLflow 0.5.1 is a patch release on top of 0.5.0 containing only bug fixes and no breaking changes or features.
Bug fixes:
- Fix
with mlflow.start_run() as run
to actually setrun
to the created Run (previously, it was None) (#322, @tomasatdatabricks) - Fixes to DBFS artifactory to throw an exception if logging an artifact fails (#309) and to mimic FileStore's behavior of logging subdirectories (#347, @andrewmchen)
- Fix for Python 3.7 support with tarfiles (#329, @tomasatdatabricks)
- Fix spark.load_model not to delete the DFS tempdir (#335, @aarondav)
- MLflow UI now appropriately shows entrypoint if it's not main (#345, @aarondav)
- Make Python API forward-compatible with newer server versions of protos (#348, @aarondav)
- Improved API docs (#305, #284, @smurching)
MLflow 0.5.0 offers some major improvements, including Keras and PyTorch first-class support as models, SFTP support as an artifactory, a new scatterplot visualization to compare runs, and a more complete Python SDK for experiment and run management.
Breaking changes:
-
The Tracking API has been split into two pieces, a "basic logging" API and a "tracking service" API. The "basic logging" API deals with logging metrics, parameters, and artifacts to the currently-active active run, and is accessible in
mlflow
(e.g.,mlflow.log_param
). The tracking service API allow managing experiments and runs (especially historical runs) and is available inmlflow.tracking
. The tracking service API will look analogous to the upcoming R and Java Tracking Service SDKs. Please be aware of the following breaking changes:-
mlflow.tracking
no longer exposes the basic logging API, onlymlflow
. So, code that was written likefrom mlflow.tracking import log_param
will have to befrom mlflow import log_param
(note that almost all examples were already doing this). -
Access to the service API goes through the
mlflow.tracking.get_service()
function, which relies on the same tracking server set by either the environment variableMLFLOW_TRACKING_URI
or by code withmlflow.tracking.set_tracking_uri()
. So code that used to look likemlflow.tracking.get_run()
will now have to domlflow.tracking.get_service().get_run()
. This does not apply to the basic logging API. -
mlflow.ActiveRun
has been converted into a lightweight wrapper aroundmlflow.entities.Run
to enable the Pythonwith
syntax. This means that there are no longer any special methods on the object returned when callingmlflow.start_run()
. These can be converted to the service API. -
The Python entities returned by the tracking service API are now accessible in
mlflow.entities
directly. Where previously you may have usedmlflow.entities.experiment.Experiment
, you would now just usemlflow.entities.Experiment
. The previous version still exists, but is deprecated and may be hidden in a future version.
-
-
REST API endpoint
/ajax-api/2.0/preview/mlflow/artifacts/get
has been moved to$static_prefix/get-artifact
. This change is coversioned in the JavaScript, so should not be noticeable unless you were calling the REST API directly (#293, @andremchen)
Features:
- [Models] Keras integration: we now support logging Keras models directly in the log_model API, model format, and serving APIs (#280, @ToonKBC)
- [Models] PyTorch integration: we now support logging PyTorch models directly in the log_model API, model format, and serving APIs (#264, @vfdev-5)
- [UI] Scatterplot added to "Compare Runs" view to help compare runs using any two metrics as the axes (#268, @ToonKBC)
- [Artifacts] SFTP artifactory store added (#260, @ToonKBC)
- [Sagemaker] Users can specify a custom VPC when deploying SageMaker models (#304, @dbczumar)
- Pyfunc serialization now includes the Python version, and warns if the major version differs (can be suppressed by using
load_pyfunc(suppress_warnings=True)
) (#230, @dbczumar) - Pyfunc serve/predict will activate conda environment stored in MLModel. This can be disabled by adding
--no-conda
tomlflow pyfunc serve
ormlflow pyfunc predict
(#225, @0wu) - Python SDK formalized in
mlflow.tracking
. This includes adding SDK methods forget_run
,list_experiments
,get_experiment
, andset_terminated
. (#299, @aarondav) mlflow run
can now be run against projects with noconda.yaml
specified. By default, an empty conda environment will be created -- previously, it would just fail. You can still pass--no-conda
to avoid entering a conda environment altogether (#218, @smurching)
Bug fixes:
- Fix numpy array serialization for int64 and other related types, allowing pyfunc to return such results (#240, @arinto)
- Fix DBFS artifactory calling
log_artifacts
with binary data (#295, @aarondav) - Fix Run Command shown in UI to reproduce a run when the original run is targeted at a subdirectory of a Git repo (#294, @adrian555)
- Filter out ubiquitious dtype/ufunc warning messages (#317, @aarondav)
- Minor bug fixes and documentation updates (#261, @stbof; #279, @dmatrix; #313, @rbang1, #320, @yassineAlouini; #321, @tomasatdatabricks; #266, #282, #289, @smurching; #267, #265, @aarondav; #256, #290, @ToonKBC; #273, #263, @mateiz; #272, #319, @adrian555; #277, @aadamson; #283, #296, @andrewmchen)
Breaking changes: None
Features:
- MLflow experiments REST API and
mlflow experiments create
now support providing--artifact-location
(#232, @aarondav) - [UI] Runs can now be sorted by columns, and added a Select All button (#227, @ToonKBC)
- Databricks File System (DBFS) artifactory support added (#226, @andrewmchen)
- databricks-cli version upgraded to >= 0.8.0 to support new DatabricksConfigProvider interface (#257, @aarondav)
Bug fixes:
- MLflow client sends REST API calls using snake_case instead of camelCase field names (#232, @aarondav)
- Minor bug fixes (#243, #242, @aarondav; #251, @javierluraschi; #245, @smurching; #252, @mateiz)
Breaking changes: None
Features:
- [Projects] MLflow will use the conda installation directory given by the
$MLFLOW_CONDA_HOME
if specified (e.g. running conda commands by invoking$MLFLOW_CONDA_HOME/bin/conda
), defaulting to running "conda" otherwise. (#231, @smurching) - [UI] Show GitHub links in the UI for projects run from http(s):// GitHub URLs (#235, @smurching)
Bug fixes:
- Fix GCSArtifactRepository issue when calling list_artifacts on a path containing nested directories (#233, @jakeret)
- Fix Spark model support when saving/loading models to/from distributed filesystems (#180, @tomasatdatabricks)
- Add missing mlflow.version import to sagemaker module (#229, @dbczumar)
- Validate metric, parameter and run IDs in file store and Python client (#224, @mateiz)
- Validate that the tracking URI is a remote URI for Databricks project runs (#234, @smurching)
- Fix bug where we'd fetch git projects at SSH URIs into a local directory with the same name as the URI, instead of into a temporary directory (#236, @smurching)
Breaking changes:
- [Projects] Removed the
use_temp_cwd
argument tomlflow.projects.run()
(--new-dir
flag in themlflow run
CLI). Runs of local projects now use the local project directory as their working directory. Git projects are still fetched into temporary directories (#215, @smurching) - [Tracking] GCS artifact storage is now a pluggable dependency (no longer installed by default).
To enable GCS support, install
google-cloud-storage
on both the client and tracking server via pip. (#202, @smurching) - [Tracking] Clients running MLflow 0.4.0 and above require a server running MLflow 0.4.0 or above, due to a fix that ensures clients no longer double-serialize JSON into strings when sending data to the server (#200, @aarondav). However, the MLflow 0.4.0 server remains backwards-compatible with older clients (#216, @aarondav)
Features:
- [Examples] Add a more advanced tracking example: using MLflow with PyTorch and TensorBoard (#203)
- [Models] H2O model support (#170, @ToonKBC)
- [Projects] Support for running projects in subdirectories of Git repos (#153, @juntai-zheng)
- [SageMaker] Support for specifying a compute specification when deploying to SageMaker (#185, @dbczumar)
- [Server] Added --static-prefix option to serve UI from a specified prefix to MLflow UI and server (#116, @andrewmchen)
- [Tracking] Azure blob storage support for artifacts (#206, @mateiz)
- [Tracking] Add support for Databricks-backed RestStore (#200, @aarondav)
- [UI] Enable productionizing frontend by adding CSRF support (#199, @aarondav)
- [UI] Update metric and parameter filters to let users control column order (#186, @mateiz)
Bug fixes:
-
Fixed incompatible file structure returned by GCSArtifactRepository (#173, @jakeret)
-
Fixed metric values going out of order on x axis (#204, @mateiz)
-
Fixed occasional hanging behavior when using the projects.run API (#193, @smurching)
-
Miscellaneous bug and documentation fixes from @aarondav, @andrewmchen, @arinto, @jakeret, @mateiz, @smurching, @stbof
Breaking changes:
- [MLflow Server] Renamed
--artifact-root
parameter to--default-artifact-root
inmlflow server
to better reflect its purpose (#165, @aarondav)
Features:
- Spark MLlib integration: we now support logging SparkML Models directly in the log_model API, model format, and serving APIs (#72, @tomasatdatabricks)
- Google Cloud Storage is now supported as an artifact storage root (#152, @bnekolny)
- Support asychronous/parallel execution of MLflow runs (#82, @smurching)
- [SageMaker] Support for deleting, updating applications deployed via SageMaker (#145, @dbczumar)
- [SageMaker] Pushing the MLflow SageMaker container now includes the MLflow version that it was published with (#124, @sueann)
- [SageMaker] Simplify parameters to SageMaker deploy by providing sane defaults (#126, @sueann)
- [UI] One-element metrics are now displayed as a bar char (#118, @cryptexis)
Bug fixes:
- Require gitpython>=2.1.0 (#98, @aarondav)
- Fixed TensorFlow model loading so that columns match the output names of the exported model (#94, @smurching)
- Fix SparkUDF when number of columns >= 10 (#97, @aarondav)
- Miscellaneous bug and documentation fixes from @emres, @dmatrix, @stbof, @gsganden, @dennyglee, @anabranch, @mikehuston, @andrewmchen, @juntai-zheng
This is a patch release fixing some smaller issues after the 0.2.0 release.
- Switch protobuf implementation to C, fixing a bug related to tensorflow/mlflow import ordering (issues #33 and #77, PR #74, @andrewmchen)
- Enable running mlflow server without git binary installed (#90, @aarondav)
- Fix Spark UDF support when running on multi-node clusters (#92, @aarondav)
-
Added
mlflow server
to provide a remote tracking server. This is akin tomlflow ui
with new options:--host
to allow binding to any ports (#27, @mdagost)--artifact-root
to allow storing artifacts at a remote location, S3 only right now (#78, @mateiz)- Server now runs behind gunicorn to allow concurrent requests to be made (#61, @mateiz)
-
TensorFlow integration: we now support logging TensorFlow Models directly in the log_model API, model format, and serving APIs (#28, @juntai-zheng)
-
Added
experiments.list_experiments
as part of experiments API (#37, @mparkhe) -
Improved support for unicode strings (#79, @smurching)
-
Diabetes progression example dataset and training code (#56, @dennyglee)
-
Miscellaneous bug and documentation fixes from @Jeffwan, @yupbank, @ndjido, @xueyumusic, @manugarri, @tomasatdatabricks, @stbof, @andyk, @andrewmchen, @jakeret, @0wu, @aarondav
- Initial version of mlflow.