W&B SDK releases

This page includes release notes for the W&B Python SDK (wandb package). For W&B Server, see W&B Server release notes.

​

v0.22.3

October 28, 2025

Added

• Settings console_chunk_max_seconds and console_chunk_max_bytes for size- and time-based multipart console logs file chunking.
• Registry API query methods (Api.registries(), Registry.{collections,versions}(), Api.registries().{collections,versions}()) now accept a per_page keyword arg to override the default batch size for paginated results.

Changed

• API keys longer than 40 characters are now supported.

Fixed

• run.config now properly returns a dict when calling artifact.logged_by() in v0.22.1.

• wandb.Api(api_key=...) now prioritizes the explicitly provided API key over thread-local cached credentials.

• Fixed a rare deadlock in console_capture.py.

  If you dump thread tracebacks during the deadlock and see the wandb-AsyncioManager-main thread stuck on a line in console_capture.py: this is now fixed.

• Fixed an issue where TensorBoard sync would sometimes stop working if the tfevents files were being written live.

• Artifact.manifest delays downloading and generating the download URL for the artifact manifest until it’s first used. If the manifest has not been locally modified, Artifact.size and Artifact.digest can return without waiting to fetch the full manifest.

• Fixed uploading GCS folder references via artifact.add_reference.

• The SDK now correctly infers notebooks paths in Jupyter sessions, using th server’s root directory, so code saving works in subdirectories (e.g. code/nested/<notebook>.ipynb).

View the v0.22.3 changelog on GitHub.

​

v0.22.2

October 07, 2025

Fixed

• Possibly fixed some cases where the output.log file was not being uploaded.
• Fixed excessive data uploads when calling run.save() repeatedly on unchanged files.

View the v0.22.2 changelog on GitHub.

​

v0.22.1

September 29, 2025

Added

• Optimize artifacts downloads re-verification with checksum caching.
• Lazy loading support for Api().runs() to improve performance when listing runs. The new lazy=True parameter (default) loads only essential metadata initially, with automatic on-demand loading of heavy fields like config and summary when accessed.
• Add storage_region option when creating artifacts. Users can use CoreWeave AI Object Storage by specifying wandb.Artifact(storage_region="coreweave-us") when using wandb.ai for faster artifact upload/download on CoreWeave’s infrastructure.

Fixed

• Api.artifact_exists() and Api.artifact_collection_exists() now raise on encountering timeout errors, rather than (potentially erroneously) returning False.

View the v0.22.1 changelog on GitHub.

​

v0.22.0

September 18, 2025

Notable Changes

This version removes support of MacOS 10.

Removed

• Remove a build targeting MacOS 10.x due to multiple security and supply chain considerations.

Fixed

• Resuming a run with a different active run will now raise an error unless you call run.finish() first, or call wandb.init() with the parameter reinit='create_new'.
• Fix Api().runs() for wandb server < 0.51.0 (when project.internalId was added to gql API).
• Sweeps: command run scripts that import readline whether directly or indirectly (e.g. import torch on Python 3.13) should no longer deadlock.

View the v0.22.0 changelog on GitHub.

​

v0.21.4

September 11, 2025

Added

• Add DSPy integration: track evaluation metrics over time, log predictions and program signature evolution to W&B Tables, and save DSPy programs as W&B Artifacts (complete program or state as JSON/PKL).

View the v0.21.4 changelog on GitHub.

​

v0.21.3

August 30, 2025

Changed

• Updated click dependency constraint from >=7.1 to >=8.0.1.

Fixed

• The message “Changes to your wandb environment variables will be ignored” is no longer printed when nothing changed.

View the v0.21.3 changelog on GitHub.

​

v0.21.2

August 28, 2025

Notable Changes

This version raises errors that would previously have been suppressed during calls to Artifact.link() or Run.link_artifact(). While this prevents undetected failures in those methods, it is also a breaking change.

Added

• New settings for max_end_of_run_history_metrics and max_end_of_run_summary_metrics.
• New wandb.integration.weave module for automatically initializing Weave when a W&B run is active and weave is imported.

Changed

• Errors encountered while linking an artifact are no longer suppressed/silenced, and Artifact.link() and Run.link_artifact() no longer return None.
• The “Run history” and “Run summary” printed at the end of a run are now limited to 10 metrics each.

Fixed

• Dataclasses in a run’s config no long raise Object of type ... is not JSON serializable when containing real classes as fields to the dataclass.
• Artifact.link() and Run.link_artifact() should be faster on server versions 0.74.0+, requiring 4-5 fewer unnecessary blocking GraphQL requests.

View the v0.21.2 changelog on GitHub.

​

v0.21.1

August 07, 2025

Notable Changes

The default ordering for Api().runs(...) and Api().sweeps(...) is now ascending order based on the runs created_at time.

Added

• Support first summary option in define_metric.
• Add support for paginated sweeps.
• pattern parameter to Api().run().files to only get files matching a given pattern from the W&B backend.
• Add optional format key to Launch input JSONSchema to specify a string with a secret format.

Changed

• Sweep.name property will now return user-edited display name if available (falling back to original name from sweep config, then sweep ID as before).
• Api().runs(...) and Api().sweeps(...) now returns runs in ascending order according to the runs created_at time.
• Artifact with large file (>2GB) uploads faster by using parallel hashing on system with more cores.
• Remove the implementation of __bool__ for the registry iterators to align with python lazy iterators.

Deprecated

• The wandb.beta.workflows module and its contents (including log_model(), use_model(), and link_model()) are deprecated and will be removed in a future release.

Fixed

• Correct the artifact url for organization registry artifacts to be independent of the artifact type.
• Suffixes on sanitized InternalArtifact names have been shortened to 6 alphanumeric characters.
• wandb.Video will not print a progress spinner while encoding video when WANDB_SILENT/WANDB_QUIET environment variables are set.
• Fixed registries fetched using api.registries() from having an extra wandb-registry- prefix in the name and full_name fields.
• Fixed a crash that could happen when using sync_tensorboard.
• Api().run(...).upload_file no longer throws an error when uploading a file in a different path relative to the provided root directory.
• Calling load() function on a public API run object no longer throws TypeError.
• When a Sweeps run function called by wandb.agent() API throws an exception, it will now appear on the logs page for the run. (This previously only happened for runs called by the wandb agent CLI command.).

View the v0.21.1 changelog on GitHub.

​

v0.21.0

July 01, 2025

Notable Changes

This version removes the legacy implementation of the service process. This is a breaking change.

Added

• Setting x_stats_track_process_tree to track process-specific metrics such as the RSS, CPU%, and thread count in use for the entire process tree, starting from x_stats_pid. This can be expensive and is disabled by default.
• Notes are now returned to the client when resuming a run.
• Added support for creating custom Vega chart presets through the API. Users can now define and upload custom chart specifications that can be then reused across runs with wandb.plot_table().

Changed

• Calling Artifact.link() no longer instantiates a throwaway placeholder run.
• wandb now attempts to use Unix sockets for IPC instead of listening on localhost, making it work in environments with more restrictive permissions (such as Databricks).
• Api.artifact() will now display a warning while fetching artifacts from migrated model registry collections.
• The .length for objects queried from wandb.Api has been deprecated. Use len(...) instead.

Removed

• Removed the legacy python implementation of the service process. The legacy-service option of wandb.require as well as the x_require_legacy_service and x_disable_setproctitle settings with the corresponding environment variables have been removed and will now raise an error if used.

• Removed the private wandb.Run._metadata attribute. To override the auto-detected CPU and GPU counts as well as the GPU type, please use the new settings x_stats_{cpu_count,cpu_logical_count,gpu_count,gpu_type}.

Fixed

• Allow s3 style CoreWeave URIs for reference artifacts.
• Fixed rare bug that made Ctrl+C ineffective after logging large amounts of data.
• Respect silent, quiet, and show_warnings settings passed to a Run instance for warnings emitted by the service process.
• api.Runs no longer makes an API call for each run loaded from W&B.
• Correctly parse the x_extra_http_headers setting from the env variable.
• .length calls the W&B backend to load the length of objects when no data has been loaded rather than returning None.

View the v0.21.0 changelog on GitHub.

​

v0.20.1

June 04, 2025

Fixed

• wandb.Image() was broken in 0.20.0 when given NumPy arrays with values in the range [0, 1], now fixed.

• wandb.Table: Added new constructor param, log_mode, with options "IMMUTABLE" and "MUTABLE". IMMUTABLE log mode (default) is existing behavior that only allows a table to be logged once. MUTABLE log mode allows the table to be logged again if it has been mutated. (@domphan-wandb in https://github.com/wandb/wandb/pull/9758)

• wandb.Table: Added a new log_mode, "INCREMENTAL", which logs newly added table data incrementally. (@domphan-wandb in https://github.com/wandb/wandb/pull/9810)

View the v0.20.1 changelog on GitHub.

​

v0.20.0

June 03, 2025

Notable Changes

This version removes the ability to disable the service process. This is a breaking change.

Added

• Added merge parameter to Artifact.add_dir to allow overwrite of previously-added artifact files.
• Support for pytorch.tensor for masks and boxes parameters when creating a wandb.Image object.
• sync_tensorboard now supports syncing tfevents files stored in S3, GCS and Azure.
  • GCS paths use the format gs://bucket/path/to/log/dir and rely on application-default credentials, which can be configured using gcloud auth application-default login
  • S3 paths use the format s3://bucket/path/to/log/dir and rely on the default credentials set through aws configure
  • Azure paths use the format az://account/container/path/to/log/dir and the az login credentials, but also require the AZURE_STORAGE_ACCOUNT and AZURE_STORAGE_KEY environment variables to be set. Some other environment variables are supported as well, see here.
• Added support for initializing some Media objects with pathlib.Path.
• New setting x_skip_transaction_log that allows to skip the transaction log. Note: Should be used with caution, as it removes the gurantees about recoverability.
• normalize parameter to wandb.Image initialization to normalize pixel values for Images initialized with a numpy array or pytorch tensor.

Changed

• Various APIs now raise TypeError instead of ValueError or other generic errors when given an argument of the wrong type.
• Various Artifacts and Automations APIs now raise CommError instead of ValueError upon encountering server errors, so as to surface the server error message.
• wandb.sdk.wandb_run.Run::save method now requires the glob_str argument.

Removed

• Removed support for disabling the service process. The x_disable_service/_disable_service setting and the WANDB_DISABLE_SERVICE/WANDB_X_DISABLE_SERVICE environment variable have been deprecated and will now raise an error if used.
• Removed ability to use wandb.docker after only importing wandb.
  • wandb.docker is not part of wandb’s public interface and is subject to breaking changes. Please do not use it.
• Removed no-op sync argument from wandb.Run::log function.
• Removed deprecated wandb.sdk.wandb_run.Run.mode property.
• Removed deprecated wandb.sdk.wandb_run.Run::join method.

Deprecated

• The start_method setting is deprecated and has no effect; it is safely ignored.
• The property Artifact.use_as and parameter use_as for run.use_artifact() are deprecated since these have not been in use for W&B Launch.

Fixed

• Calling wandb.teardown() in a child of a process that called wandb.setup() no longer raises WandbServiceNotOwnedError.
  • This error could have manifested when using W&B Sweeps
• Offline runs with requested branching (fork or rewind) sync correctly.
• Log exception as string when raising exception in Job wait_until_running method.
• wandb.Image initialized with tensorflow data would be normalized differently than when initialized with a numpy array.
• Using wandb login no longer prints a warning about wandb.require("legacy-service").
• Logging a Table (or other objects that create internal artifacts) no longer raises ValueError when logged from a run whose ID contains special characters.
• wandb.Api initialized with the base_url now respects the provided url, rather than the last login url.

View the v0.20.0 changelog on GitHub.

​

v0.19.11

May 07, 2025

Added

• Added creation, deletion, and updating of registries in the SDK.
• artifact.is_link property to artifacts to determine if an artifact is a link artifact (such as in the Registry) or source artifact.
• artifact.linked_artifacts to fetch all the linked artifacts to a source artifact and artifact.source_artifact to fetch the source artifact of a linked artifact.
• run.link_artifact(), artifact.link(), and run.link_model() all return the linked artifact upon linking.
• Multipart download for artifact file larger than 2GB, user can control it directly using artifact.download(multipart=True).
• Project.id property to get the project ID on a wandb.public.Project.
• New public API for W&B Automations.
  • New submodules and classes in wandb.automations.* to support programmatically managing W&B Automations.
  • Api.integrations(), Api.slack_integrations(), Api.webhook_integrations() to fetch a team’s existing Slack or webhook integrations.
  • Api.create_automation(), Api.automation()/Api.automations(), Api.update_automation(), Api.delete_automation() to create, fetch, edit, and delete Automations.
• Create and edit automations triggered on RUN_METRIC_CHANGE events, i.e. on changes in run metric values (absolute or relative deltas).
• Ability to collect profiling metrics for Nvidia GPUs using DCGM. To enable, set the WANDB_ENABLE_DCGM_PROFILING environment variable to true. Requires the nvidia-dcgm service to be running on the machine. Enabling this feature can lead to increased resource usage.

Fixed

• run.log_code correctly sets the run configs code_path value.
• Correctly use WANDB_CONFIG_DIR for determining system settings file path.
• Prevent invalid Artifact and ArtifactCollection names (which would make them unloggable), explicitly raising a ValueError when attempting to assign an invalid name.
• Prevent pydantic ConfigError in Pydantic v1 environments from not calling .model_rebuild()/.update_forward_refs() on generated types with ForwardRef fields.
• wandb.init() no longer raises Permission denied error when the wandb directory is not writable or readable.
• Calling file.delete() on files queried via api.Runs(...) no longer raises CommError.
  • Bug introduced in 0.19.1

View the v0.19.11 changelog on GitHub.

​

v0.19.10

April 22, 2025

Added

• The new reinit="create_new" setting causes wandb.init() to create a new run even if other runs are active, without finishing the other runs (in contrast to reinit="finish_previous"). This will eventually become the default.
• Added Artifact.history_step to return the nearest run step at which history metrics were logged for the artifact’s source run.
• Added data_is_not_path flag to skip file checks when initializing wandb.Html with a sting that points to a file.

Changed

• Artifact.download() no longer raises an error when using WANDB_MODE=offline or when an offline run exists.

Removed

• Dropped the -q / --quiet argument to the wandb magic in IPython / Jupyter; use the quiet run setting instead.

Deprecated

• The following wandb.Run methods are deprecated in favor of properties and will be removed in a future release (@kptkin in https://github.com/wandb/wandb/pull/8925):.
  • run.project_name() is deprecated in favor of run.project
  • run.get_url() method is deprecated in favor of run.url
  • run.get_project_url() method is deprecated in favor of run.project_url
  • run.get_sweep_url() method is deprecated in favor of run.sweep_url

Fixed

• Fixed ValueError on Windows when running a W&B script from a different drive.
• Fix base_url setting was not provided to wandb.login.
• wandb.Html() no longer raises IsADirectoryError with a value that matched a directory on the users system.

View the v0.19.10 changelog on GitHub.

​

v0.19.9

April 01, 2025

Added

• The reinit setting can be set to "default".
• Added support for building artifact file download urls using the new url scheme, with artifact collection membership context.

Changed

• Boolean values for the reinit setting are deprecated; use “return_previous” and “finish_previous” instead.
• The “wandb” logger is configured with propagate=False at import time, whereas it previously happened when starting a run. This may change the messages observed by the root logger in some workflows.
• Metaflow now requires plum-dispatch package.
• Relaxed the pydantic version requirement to support both v1 and v2.
• Existing pydantic types have been adapted to be compatible with Pydantic v1.
• wandb.init(dir=...) now creates any nonexistent directories in dir if it has a parent directory that is writeable.
• The server now supports fetching artifact files by providing additional collection information; updated the artifacts api to use the new endpoints instead.
• Paginated methods (and underlying paginators) that accept a per_page argument now only accept int values. Default per_page values are set directly in method signatures, and explicitly passing None is no longer supported.

Fixed

• Calling wandb.init() in a notebook finishes previous runs as previously documented.
  • Bug introduced in 0.19.0
• Fixed an error being thrown when logging jpg/jpeg images containing transparency data.
• wandb.init(resume_from=...) now works without explicitly specifying the run’s id.
• Deleting files with the Public API works again.
  • Bug introduced in 0.19.1
• Fixed media files not displaying in the UI when logging to a run with a custom storage bucket.

View the v0.19.9 changelog on GitHub.

​

v0.19.8

March 04, 2025

Fixed

• Media file paths containing special characters (?, *, ], [ or \) no longer cause file uploads to fail in wandb-core.

Changed

• The system monitor now samples metrics every 15 seconds by default, up from 10 seconds.

View the v0.19.8 changelog on GitHub.

​

v0.19.7

February 21, 2025

Added

• Registry search api.

Changed

• Changed moviepy constraint to >=1.0.0.
• wandb.init() displays more detailed information, in particular when it is stuck retrying HTTP errors.

Removed

• Removed the private x_show_operation_stats setting.

Fixed

• Fixed incorrect logging of an “wandb.Video requires moviepy […]” exception when using moviepy v2.
• wandb.setup() correctly starts up the internal service process; this semantic was unintentionally broken in 0.19.2.
• Fixed TypeError: Object of type ... is not JSON serializable when using numpy number types as values.

View the v0.19.7 changelog on GitHub.

​

v0.19.6

February 05, 2025

Added

• Prometheus API support for Nvidia DCGM GPU metrics collection.

Changed

• Changed Nvidia GPU ECC counters from aggregated to volatile.

Fixed

• Fixed a performance issue causing slow instantiation of wandb.Artifact, which in turn slowed down fetching artifacts in various API methods.
• Some errors from wandb.Api have better string representations.
• Artifact.add_reference, when used with file URIs for a directory and the name parameter, was incorrectly adding the value of name to the path of the file references.
• Fixed a bug causing Artifact.add_reference() with checksum=False to log new versions of local reference artifacts without changes to the reference URI.

View the v0.19.6 changelog on GitHub.

​

v0.19.5

January 29, 2025

Added

• Added wandb login --base-url {host_url} to login as an alias of wandb login --host {host_url}.

Changed

• Temporarily disabled collecting per-core CPU utilization stats.

Fixed

• Fixed a bug causing offline mode to make network requests when logging media artifacts. If you are using an older version of W&B Server that does not support offline artifact uploads, use the setting allow_offline_artifacts=False to revert to older compatible behavior.
• Expand sanitization rules for logged table artifact name to allow for hyphens and dots. This update brings the rules up-to-date with the current rules for artifact names. (Allowing letters, numbers, underscores, hyphens, and dots).
• Correctly handle run rewind settings fork_from and resume_from.

View the v0.19.5 changelog on GitHub.

​

v0.19.4

January 16, 2025

Fixed

• Fix incorrectly reported device counts and duty cycle measurements for TPUs with single devices per chip / multiple devices on the host and make TPU metrics sampling more robust.
• Handle non-consecutive TPU device IDs in system monitor.

View the v0.19.4 changelog on GitHub.

​

v0.19.3

January 13, 2025

Fixed

• Fix wandb.Settings update regression in wandb.integration.metaflow.

View the v0.19.3 changelog on GitHub.

​

v0.19.2

January 07, 2025

Added

• Support JWT authentication in wandb-core.
• Add support for logging nested custom charts.

Changed

• Calling wandb.init(mode="disabled") no longer disables all later runs by default. Use wandb.setup(settings=wandb.Settings(mode="disabled")) for this instead, or set mode="disabled" explicitly in each call to wandb.init().

Fixed

• The stop button correctly interrupts runs whose main Python thread is running C code, sleeping, etc.
• Remove unintentional print that occurs when inspecting wandb.Api().runs().
• Fix uploading large artifacts when using Azure Blob Storage.
• The wandb offline command no longer adds an unsupported setting to wandb.Settings, resolving ValidationError.
• Fix error when reinitializing a run, caused by accessing a removed attribute.
• Fixed occasional deadlock when using multiprocessing to update a single run from multiple processes.
• Prevent errors from bugs in older versions of botocore < 1.5.76.
• Fixed various checks against invalid anonymous settings value.

Removed

• The wandb.wandb_sdk.wandb_setup._setup() function’s reset parameter has been removed. Note that this is a private function, even though there appear to be usages outside of the repo. Please wandb.teardown() instead of _setup(reset=True).
• In the private wandb.wandb_sdk.wandb_setup module, the logger and _set_logger symbols have been removed.

Security

• Bump github.com/go-git/go-git version to 5.13.0 to address CVE-2025-21613.
• Bump golang.org/x/net version to 0.33.0 to address CVE-2024-45338.

View the v0.19.2 changelog on GitHub.

​

v0.19.1

December 13, 2024

Fixed

• Fixed bug where setting WANDB__SERVICE_WAIT led to an exception during wandb.init.

Changed

• run.finish() displays more detailed information in the terminal and in Jupyter notebooks.
• Improved error message for failing tensorboard.patch() calls to show the option to call tensorboard.unpatch() first.
• Add projectId to deleteFiles mutation if the server supports it.

Security

• Bump golang.org/x/crypto version to 0.31.0 to address CVE-2024-45337.

View the v0.19.1 changelog on GitHub.

​

v0.19.0

December 05, 2024

Notable Changes

This version drops Python 3.7 and removes the wandb.Run.plot_table method. This version adds pydantic>=2.6,<33 as a dependency.

Changed

• Set default behavior to not create a W&B Job.
• Add pydantic>=2.6,<33 as a dependency (@dmitryduev in https://github.com/wandb/wandb/pull/8649 & estellazx. in https://github.com/wandb/wandb/pull/8905)

Removed

• Remove wandb.Run.plot_table method. The functionality is still available and should be accessed using wandb.plot_table, which is now the recommended way to use this feature.
• Drop support for Python 3.7.

Fixed

• Fix ultralytics reporting if there are no positive examples in a validation batch.
• Debug printing for hyperband stopping algorithm printed one char per line.
• Include the missing log_params argument when calling lightgbm’s wandb_callback function.

View the v0.19.0 changelog on GitHub.

​

v0.18.7

November 13, 2024

Added

• Added create_and_run_agent to __all__ in wandb/sdk/launch/__init__.py to expose it as a public API.

Changed

• Tables logged in offline mode now have updated keys to artifact paths when syncing. To revert to old behavior, use setting allow_offline_artifacts = False.

Deprecated

• The quiet argument to wandb.run.finish() is deprecated, use wandb.Settings(quiet=...) to set this instead.

Fixed

• Fix api.artifact() to correctly pass the enable_tracking argument to the Artifact._from_name() method.

View the v0.18.7 changelog on GitHub.

​

v0.18.6

November 06, 2024

Added

• Add a boolean overwrite param to Artifact.add()/Artifact.add_file() to allow overwrite of previously-added artifact files.

Fixed

• Add missing type hints of the wandb.plot module in the package stub.
• Fix limiting azure reference artifact uploads to max_objects.
• Fix downloading azure reference artifacts with skip_cache=True.
• Fix multipart uploads for files with no content type defined in headers.
• Fixed tensorboard failing to sync when logging batches of images.
• Fixed behavior of mode='x'/mode='w' in Artifact.new_file() to conform to Python’s built-in file modes.
• Do not ignore parameter distribution when configuring sweep parameters from SDK.

Changed

• Added internal method, api._artifact(), to fetch artifacts so that usage events are not created if not called by an external user.
• Changed default mode in Artifact.new_file() from 'w' to 'x' to accurately reflect existing default behavior.

View the v0.18.6 changelog on GitHub.

​

v0.18.5

October 17, 2024

Fixed

• Import Literal from typing_extensions in Python 3.7; broken in 0.18.4.

View the v0.18.5 changelog on GitHub.

​

v0.18.4

October 17, 2024

Added

• Track detailed metrics for Apple ARM systems including GPU, eCPU, and pCPU utilization, power consumption, and temperature, and memory/swap utilization.
• Allow users to link Registry artifacts without inputting the organization entity name.
• Added a warning message indicating that the fps argument will be ignored when creating a wandb.Video object from a file path string or a bytes object.
• Update docstrings for logged_artifacts and used_artifacts methods in Run class.
• The _show_operation_stats setting enables a preview of a better run.finish() UX.

Fixed

• Log power on AMD MI300X series GPUs.
• Fixed typing issue of wandb.Api.
• Ensure artifact objects are fully updated on Artifact.save().

Changed

• Updated minimum version of sentry-sdk to 2.0.0 to address deprecation warnings.

View the v0.18.4 changelog on GitHub.

​

v0.18.3

October 01, 2024

Added

• Add the ability to monitor the utilization metrics of Google’s Cloud TPU devices.

Fixed

• Capture Nvidia GPU stats on Windows.
• Fixed a regression introduced in v0.18.2 that affected capturing the names of Nvidia GPU devices.
• run.log_artifact() no longer blocks other data uploads until the artifact upload finishes.
• Fixed media dependency for rdkit updated from rdkit-pypi to rdkit.
• Saving an artifact with many large files no longer exhausts OS threads.

Changed

• After artifact = run.log_artifact(), you must use artifact.wait() before operations that rely on the artifact having been uploaded. Previously, this wasn’t necessary in some cases because run.log_artifact() blocked other operations on the run.

View the v0.18.3 changelog on GitHub.

​

v0.18.2

September 27, 2024

Added

• Add upsert_run_queue method to wandb.Api.
• Add tags parameter to wandb.Api.artifacts() to filter artifacts by tag.

Fixed

• Update the signature and docstring of wandb.api.public.runs.Run.log_artifact() to support artifact tags like Run instances returned by wandb.init().
• Add docstring for wandb.watch to support auto-complete.
• Fix glob matching in define metric to work with logged keys containing /.
• Allow a\.b syntax in run.define_metric to refer to a dotted metric name.
  • NOTE: Not fixed if using wandb.require("legacy-service")
• Fix Unknown image format error when uploading a gif through tensorboard.
• Fix OSError from calling Artifact.add_file with file paths on mounted filesystems.
• Restored compatibility for macOS versions <== 10.15 for wandb-core.

View the v0.18.2 changelog on GitHub.

​

v0.18.1

September 16, 2024

Fixed

• Allow all users to read cache files when core is enabled.
• Infinite scalars logged in TensorBoard are uploaded successfully rather than skipped.
• Properly respect WANDB_ERROR_REPORTING=false. This fixes a regression introduced in 0.18.0.

Changed

• Remove sentry logging for sendLinkArtifact.
• Default to capturing requirements.txt in Run.log_code.

View the v0.18.1 changelog on GitHub.

​

v0.18.0

September 11, 2024

Notable Changes

This version switches wandb to a new backend by enabling wandb.require("core") by default. This should not be a breaking change, but the new backend may have unexpected differences in behavior for legacy functionality and rare edge cases.

Added

• Add support for artifact tags, via Artifact.tags and Run.log_artifact().

Fixed

• Detect the notebook name in VS Code’s built-in jupyter server.

Changed

• The new “core” backend, previously activated using wandb.require(“core”), is now used by default. To revert to the legacy behavior, add wandb.require("legacy-service") at the beginning of your script. Note: In a future minor release, the option to disable this new behavior will be removed.

View the v0.18.0 changelog on GitHub.

​

v0.17.9

September 05, 2024

Changed

• Changed the default system metrics sampling interval to 10 seconds without averaging, while allowing custom intervals via wandb.init(settings=wandb.Settings(_stats_sampling_interval=...)).

Deprecated

• define_metric(summary='best', goal=...) is deprecated and soon will be removed, use define_metric(summary='min') or define_metric(summary='min') instead.

View the v0.17.9 changelog on GitHub.

​

v0.17.8

August 28, 2024

Added

• Capture SM (Streaming Multiprocessor), memory, and graphics clock speed (MHz), (un)corrected error counts, fan speed (%), and encoder utilization for Nvidia GPU devices when using core.
• Allow iterating over wandb.Config like a dictionary.
• PR curves, images and histograms are supported when using TensorBoard with core enabled.
• Added wandb.require("legacy-service") as the opt-out analog of wandb.require("core").

Fixed

• Avoid leaving behind wandb-core process if user hits Ctrl+C twice.
• Fix deprecation warnings arising from NumPy >= 2.1 by removing newshape argument from numpy.reshape.
• Skip uploading/downloading GCS reference artifact manifest entries corresponding to folders.

Deprecated

• Ability to disable the service process (WANDB__DISABLE_SERVICE) is deprecated and will be removed in the next minor release.

View the v0.17.8 changelog on GitHub.

​

v0.17.7

August 15, 2024

Fixed

• Ensure Nvidia GPU metrics are captured if libnvidia-ml.so is unavailable when using core.
• Allow define_metric("x", step_metric="x") when using core.
• Correctly upload empty files when using core.
• Fix occasional “send on closed channel” panic when finishing a run using core.

View the v0.17.7 changelog on GitHub.

​

v0.17.6

August 08, 2024

Added

• Specify job input schemas when calling manage_config_file or manage_wandb_config to create a nicer UI when launching the job.
• Use the filesystem rather than protobuf messages to transport manifests with more than 100k entries to the core process.
• Adds the box3d constructor for Box3D.

Changed

• run.define_metric() raises an error when given extraneous arguments.
• In disabled mode, use the wandb.sdk.wandb_run.Run class instead of wandb.sdk.wandb_run.RunDisabled, which has been removed.
• When WANDB_MODE = offline calling artifact.download() now throws an error instead of stalling.

Fixed

• Correctly handle directory stubs when logging external artifact in azure storage account with Hierarchical Namespace enabled.
• Docstring in api.runs() regarding default sort order, missed in https://github.com/wandb/wandb/pull/7675.

View the v0.17.6 changelog on GitHub.

​

v0.17.5

July 19, 2024

Added

• When using wandb-core, support multipart uploads to S3.

Changed

• run.finish() may raise an exception in cases where previously it would os._exit().
• run.link_artifact() can now surface server errors.

Fixed

• Handle path_prefixes that don’t correspond to directory names when downloading artifacts.
• Fix creating or updating an artifact with the incremental=True flag.
• Use filled resource_arg macros when submitting W&B Launch jobs to AmazonSageMaker.

View the v0.17.5 changelog on GitHub.

​

v0.17.4

July 03, 2024

Added

• Support queue template variables in launch sweep scheduler jobs.

Fixed

• Use sys.exit() instead of os._exit() if an internal subprocess exits with a non-zero code.
• Fix an occasional race condition when using core that could affect run logs.
• Fix OSError on Artifact.download(skip_cache=True) when encountering different filesystems.

View the v0.17.4 changelog on GitHub.

​

v0.17.3

June 24, 2024

Fixed

• Correctly name the netrc file on Windows as _netrc.
• With core enabled, nested tqdm bars show up correctly in the Logs tab.
• Fix W&B Launch registry ECR regex separating tag on forward slash and period.

View the v0.17.3 changelog on GitHub.

​

v0.17.2

June 17, 2024

Added

• Add prior runs when creating a sweep from the CLI.

Fixed

• Fix issues with numpy>=2 support.
• Fix “UnicodeDecodeError: ‘charmap’” when opening HTML files on Windows by specifying UTF-8 encoding.
• Ensure Artifact.delete() on linked artifacts only removes link, not source artifact.
• Sweep runs no longer appear to be resumed when they are not by @TimH98 https://github.com/wandb/wandb/pull/7684.

Changed

• Upgrade github.com/vektah/gqlparser/v2 from 2.5.11 to 2.5.16.

View the v0.17.2 changelog on GitHub.

​

v0.17.1

June 07, 2024

Added

• Added api.runs().histories() to fetch history metrics for runs that meet specified conditions.
• Display warning when Kubernetes pod fails to schedule.
• Added ArtifactCollection.save() to allow persisting changes.
• Added the ability to overwrite history of previous runs at an arbitrary step and continue logging from that step.
• Added new Workspace API for programatically editing W&B Workspaces.
• Added Artifact.unlink() to allow programmatic unlinking of artifacts.
• Added basic TensorBoard support when running with wandb.require("core").
  • The TensorBoard tab in W&B will work.
  • Charts show up in W&B, possibly better than when running without core.
  • Not all types of data are supported yet. Unsupported data is not shown in charts.

Fixed

• Fix define_metric behavior for multiple metrics in shared mode.
• Correctly pass in project name to internal api from run while calling run.use_artifact().
• Correctly upload console output log files when resuming runs enabled with console_multipart setting by @kptkin in https://github.com/wandb/wandb/pull/7694 and @dmitryduev in https://github.com/wandb/wandb/pull/7697.
• Interpret non-octal strings with leading zeros as strings and not integers in sweep configs by @KyleGoyette https://github.com/wandb/wandb/pull/7649.
• Support Azure repo URI format in Launch @KyleGoyette https://github.com/wandb/wandb/pull/7664.
• Fix path parsing for artifacts with aliases containing forward slashes.
• Add missing docstrings for any public methods in Api class.
• Correctly add latest alias to jobs built by the job builder @KyleGoyette https://github.com/wandb/wandb/pull/7727.

Changed

• Option to change naming scheme for console output logs from output.log to logs/YYYYMMDD_HHmmss.ms_output.log.
• Require unsafe=True in use_model calls that could potentially load and deserialize unsafe pickle files by @anandwandb https://github.com/wandb/wandb/pull/7663.
• Update order in api.runs() to ascending to prevent duplicate responses by @thanos-wandb https://github.com/wandb/wandb/pull/7675.
• Eliminate signed URL timeout errors during artifact file uploads in core.

Deprecated

• Deprecated ArtifactCollection.change_type() in favor of ArtifactCollection.save().

View the v0.17.1 changelog on GitHub.

​

v0.17.0

May 07, 2024

Notable Changes

Renamed wandb.plots to wandb.plot, renamed all integrations from wandb.<name> to wandb.integration.<name>, and removed the [async] extra.

This version packages the wandb-core binary, formerly installed by the wandb-core Python package on PyPI. The wandb-core package is now unused and can be uninstalled.

Added

• The wandb package now includes the wandb-core binary.
  • wandb-core is a new and improved backend for the W&B SDK that focuses on performance, versatility, and robustness.
  • Currently, it is opt-in. To start using the new backend, add wandb.require("core") to your script after importing wandb.
• wandb-core now supports Artifact file caching by @moredatarequired in https://github.com/wandb/wandb/pull/7364 and https://github.com/wandb/wandb/pull/7366.
• Added artifact_exists() and artifact_collection_exists() methods to Api to check if an artifact or collection exists.
• wandb launch -u <git-uri | local-path> creates and launches a job from the given source code.

Fixed

• Prevent crash on run.summary for finished runs.
• Correctly report file upload errors when using wandb-core.
• Implemented a stricter check for AMD GPU availability.
• Fixed run.save() on Windows.
• Show a warning instead of failing when using registries other than ECR and GAR with the Kaniko builder.
• Fixed wandb.init() type signature including None.

Changed

• When using wandb-core need to specify a required flag (wandb.require("core")) to enable it, before it was picked up automatically.
• Use ETags instead of MD5 hashes for GCS reference artifacts.

Removed

• Removed the deprecated wandb.plots.* functions and top-level third-party integrations wandb.[catboost,fastai,keras,lightgbm,sacred,xgboost]. Please use wandb.plot instead of wandb.plots and wandb.integration.[catboost,fastai,keras,lightgbm,sacred,xgboost] instead of wandb.[catboost,fastai,keras,lightgbm,sacred,xgboost].
• Removed the [async] extra and the _async_upload_concurrency_limit setting.
• Removed undocumented settings: _except_exit and problem.

View the v0.17.0 changelog on GitHub.

​

v0.16.6

April 03, 2024

Added

• Added support for overriding kaniko builder settings in the agent config.
• Added link to the project workspace of a run in the footer.
• Added support for overriding stopped run grace period in the agent config.
• Added setting (_disable_update_check) to disable version checks during init.
• WandbLogger.sync in the OpenAI Fine-Tuning integration gets a new log_datasets boolean argument to turn off automatic logging of datasets to Artifacts.
• Reduced default status print frequency of launch agent. Added verbosity controls to allow for increased status print frequency and printing debug information to stdout.

Changed

• Limit policy option on artifact cli’s put() to choices, [“mutable”, “immutable”].
• Updated artifact public api methods to handle nullable Project field on the ArtifactSequence/ArtifactCollection type, based on gorilla server changes.

Fixed

• Fixed run.save() not working with files inside run.dir, introduced in previous release.
• Fixed rare panic during large artifact uploads.
• Fixed wandb.login causing runs not to be associated with launch queue.
• Fixed job artifact download failing silently and causing run crash when using W&B Launch by @KyleGoyette https://github.com/wandb/wandb/pull/7285.
• Fix handling of saving training files to Artifacts in the OpenAI Fine-Tuning integration.

View the v0.16.6 changelog on GitHub.

​

v0.16.5

March 25, 2024

Added

• Added feature to move staging files to cache (instead of copying) for mutable artifact file uploads when caching is enabled.
• Added support to skip caching files to the local filesystem while uploading files to artifacts.
• Added support to skip staging artifact files during upload by selecting a storage policy.
• Preliminary support for forking a run using wandb.init(fork_from=...).
• run.save() accepts pathlib.Path values;.

Changed

• When printing the run link point to the workspace explicitly.

Fixed

• In case of transient server issues when creating the wandb API key kubernetes secret, we’ll retry up to 5 times.

Removed

• When printing run’s information in the terminal remove links to jobs.

View the v0.16.5 changelog on GitHub.

​

v0.16.4

March 05, 2024

Added

• Added ability to change artifact collection types.
• Add support for installing deps from pyproject.toml.
• Support kaniko build with user-provided pvc and docker config.
• Added ability to import runs between W&B instances.

Changed

• Wandb-core rate-limits requests to the backend and respects RateLimit-* headers. by @timoffex in https://github.com/wandb/wandb/pull/7065

Fixed

• Fix passing of template variables in the sweeps-on-launch scheduler.
• Link job artifact to a run to be specified as input.
• Fix sagemaker entrypoint to use given entrypoint.
• Parse upload headers correctly.
• Properly propagate server errors.
• Make file upload faster by using parallelism.
• Don’t send git data if it’s not populated.
• Fix console logging resumption, avoid overwrite.
• Remove hostname validation when using —host on wandb login.
• Don’t discard past visualizations when resuming a run.
• Avoid retrying on conflict status code.
• Fix visualization config merging for resumed runs in wandb-core.
• Replace usage of standard library’s json with segmentio’s.
• Remove stderr as writer for the logs.
• Disable negative steps from initialization.
• Fix report loading in pydantic26.
• Revert “make upload request async to support progress reporting (#6497)”.
• Fix entrypoint specification when using a Dockerfile.wandb.
• Fix stream releasing probe handle too early.
• Always attempt to pull latest image for local container.

View the v0.16.4 changelog on GitHub.

​

v0.16.3

February 06, 2024

Enhancements

• Generate data type info in core.
• Add support for Launch.
• Added option to control number of grouped sampled runs in reports.
• Add shared mode to enable multiple independent writers to the same run.
• Reduce artifact download latency via optional cache copy + threads.
• Add partial file downloads, via directory prefix.
• Update the Diffusers Integration.
• Update Ultralytics Integration.
• Add Pytorch Lightning Fabric Logger.
• Update go packages.

Fixes

• Remove project and runner fields from agent config.
• Recognize deleted k8s jobs as failed.
• Warn of extra fields in environment block instead of erroring.
• Entity override bug where ENVVAR is prioritized over kwargs.
• Local container runner doesn’t ignore override args.
• Merge-update config with sweep/launch config.
• Fix retry logic in wandb-core and system_tests conftest.
• Use RW locks in system monitor’s assets management.
• Set build context to entrypoint dir if it contains Dockerfile.wandb.
• Security(launch): Mount wandb api key in launch job pods from a k8s secret.
• Wandb job create should not look for requirements.txt if Dockerfile.wandb is next to entrypoint.
• Fix _parse_path when only id is passed to wandb.Api().run().
• Update video.py: Fix fps bug.
• Clean up temp folders.
• Fix long artifact paths on Windows.
• Update Report API to work with pydantic2.6.
• Fetch all commits to enable checking out by sha.
• Dont swallow exceptions in pyagent.
• Artifact file upload progress in nexus.
• Exercise caution in system monitor when rocm-smi is installed on a system with no amd gpus.
• Typo in cli.py.
• Remove deadsnakes from accelerator build step.

View the v0.16.3 changelog on GitHub.

​

v0.16.2

January 09, 2024

Enhancements

• Refactor store logic and add store writer.
• Add AMD GPU monitoring.
• Add console log file upload.
• Add registry uri field to builders.
• Add wandb beta sync feature to upload runs to W&B.
• CLI supports allow-listed queue parameters.
• Add support for requirements and patch.diff.
• Capture SLURM-related env vars in metadata.
• —priority flag on wandb launch command to specify priority when enqueuing jobs.
• Add verify feature to wandb login.
• Sweeps on Launch honors selected job priority for sweep runs.
• Commence operation SDKrab.
• Make upload request async to support progress reporting.
• Add TensorBoard log dir watcher.
• Upload wandb-summary.json and config.yaml files.

Fixes

• Update error message and remove extra.
• Clean up issues with file sending.
• Add jitter to retry clients’ backoff strategy.
• Only skip file download if digest matches.
• Fix resume and add tests.
• Capture errors from the creation of k8s job from yaml.
• Get default entity before checking template vars.
• Remove run creation for artifact downloads if not using core.
• Retrieve ETag for ObjectVersion instead of Object for versioned buckets.
• Revert #6759 and read object version etag in place.
• Put back the upload file count.
• Add send cancel request to sender.
• Check errors in memory monitoring.
• Add job_type flag in CLI to allow override of job_type.
• Handle ultralytics utils import refactor.
• Download artifacts with wandb-core without an active run.
• Fix error logging matplotlib scatter plot if the minimum version of plotly library is not met.
• Allow overriding the default .netrc location with NETRC env var.

View the v0.16.2 changelog on GitHub.

​

v0.16.1

December 05, 2023

Enhancements

• Remove recursive download.
• Add debounce summary in handler.
• Fix bug in ultralytics import and version pinning.
• Support template variables when queueing launch runs.
• Add —skip-console option to offline sync cli command.
• Add basic graphql versioning mechanism.
• Add Apple M* GPU stats monitoring.
• Add helper to load wandb.Config from env vars.
• Port OpenAI WandbLogger for openai-python v1.0.
• Fix version check for openAI WandbLogger.
• Diffusers autologger.
• Adding parameter to image to specify file type jpg, png, bmp, gif.

Fixes

• Make offline sync work properly.
• Fix run existence check to not depend on files being uploaded in the run.
• Gcp storage uri verifaction failed due to improper async wrapping.
• Updating summary with nested dicts now doesn’t throw an error.
• Add prioritization mode to RunQueue create.
• Create symlink to the server logs in the runs folder.
• Single value problem in wandb/wandb_torch.py::log_tensor_stats.
• Tmin vs tmax in wandb/wandb_torch.py::log_tensor_stats.
• Minor fix up for non server cases.
• Make old settings more robust.
• Handle tags when resuming a run.

View the v0.16.1 changelog on GitHub.

​

v0.16.0

November 07, 2023

Enhancements

• Add nested config support.
• Finish artifact saver.
• Add sync file counts in the footer.
• Implement directory watcher and related functionality.
• Run agent on an event loop.
• Add sampled history.
• Prototype for models api.
• Register sweep scheduler virtual agents async.
• Make file uploads work with local.
• Allow users to log custom chart tables in a different section.
• Generalize uploader to filemanager to allow downloads.
• Drop python 3.6 support.
• Delete staging files in Nexus.
• Set up artifact downloads to use sdk nexus core.
• Add file upload progress and make completion callback a list of callbacks.
• Add wandb.ai/run-id label to jobs.

Fixes

• Ensures that complete run.config is captured in SageMaker.
• Improve error handling for gitlib for FileNotFoundErrors.
• Increase max message size, handle errors.
• Agent better balances multiple queues.
• Remove versioning enabled check in GCS reference handler.
• Add google-cloud-aiplatform to launch shard.
• Add use_artifact as passthrough messages.
• Adjust vertex environment variables.
• Update artifacts cache file permissions for NamedTemporaryFile.
• Fix a number of issues.
• Saving mlp figure to buffer and reading with PIL should specify format.
• Send content-length, check response status code in uploader.
• Fix step logic when resuming runs with no metrics logged.
• Hook_handle being set to list instead of dict on unhook.
• Verify gcp credentials before creating vertex job.
• Add load option to docker buildx.
• Fix perf regressions in artifact downloads and fix file download location.
• Add warning when log_code can’t locate any files.
• Adjust ipython hooks for v8.17.

View the v0.16.0 changelog on GitHub.

​

v0.15.12

October 03, 2023

Enhancements

• Implement config debouncing mechanism.
• Fix channel swapping on ultrlytics classification task.
• Implement nexus alpha cpp interface.
• Expose system metrics in the run object (PoC).
• Pin ultralytics version support to v8.0.186.

Fixes

• Get logs from failed k8s pods.
• Allow adding s3 bucket as reference artifact.
• Fix race condition in agent thread clean up.
• Don’t assume run and its i/o artifacts are in the same project.
• Fix wandb.Api().run(run_name).log_artifact(artifact).
• ValueError with None value in sweep.
• Fix typo in s3 handler.
• Fix the argument order for new argument target_fraction.
• Fix potential race in config debouncer.
• Fix graphql type mapping.
• Fix concurrency limit in uploader.

View the v0.15.12 changelog on GitHub.

​

v0.15.11

September 21, 2023

Enhancements

• Add support for code saving in script mode.
• Add support for policy=end in wandb.save.
• Add system info to metadata.
• Add nvidia gpu system info to metadata.
• Delete run queues with public api.
• Introduce custom proxy support for wandb http(s) traffic.

Fixes

• Fix logger when logging filestream exception.
• Use watch api to monitor launched CRDs.
• Forbid enqueuing docker images without target project.
• Add missing Twitter import for API users.
• Get S3 versionIDs from directory references.
• Make watch streams recover from connection reset.
• Use json.loads(…, strict=False) to ignore invalid utf-8 and control characters in api.Run.load.
• Correctly identify colab as a jupyter-like env in settings.
• Improve memory management for summary updates.
• Add environment variable to configure batch size for download urls.
• Fail rqis if no run is created.

Cleanup

• Update README.md.

View the v0.15.11 changelog on GitHub.

​

v0.15.10

September 06, 2023

Enhancements

• Add async support to Autologger API and enable it for Openai.
• Add official support for python 3.11 and drop support for python 3.6.
• Implement network logging and file pusher timeout for debugging.
• Set ttl(time to live) for artifact versions.
• Add support for define metric.
• Include agent version when creating launch agent.
• Runless git jobs can use requirements.txt in parent directories.
• Retrieve the parent collection from an Artifact.
• Improve file uploads.
• Add run id option to artifact put method to log artifacts to existing runs.
• Add metadata property to Run object.
• Support setting a custom Dockerfile in launch overrides.
• Add Nvidia GPU asset to system monitor.
• Enable deleting artifact collections from SDK.
• Add dockerfile CLI param & use Dockerfile.wandb by default if present.
• Extend cache cleanup to allow specifying a target fraction.
• Add an eval-able repr to ArtifactManifestEntry.
• Enable docker-based wheel building for nexus.
• Add Nvidia GPU asset to system monitor.
• Clear the cache on add to prevent overflow.
• Capture disk i/o utilization in system metrics.
• Add disk io counters to monitor metrics.
• Make paths for disk usage monitoring configurable.
• Use WANDB_SWEEP_ID to include a run in an existing sweep.
• Handle LinkArtifact calls made to Nexus Core.
• Fix retry logic for http clients and allow user customization.
• Support user defined headers in the gql client transport.
• Enable set types in wandb.Config.
• Visualize images with bbox overlays for ultralytics.
• Add exponential decay sampling utility for line_plot.
• Always print the traceback inside of the wandb.init context manager.
• Add setting to disable automatic machine info capture.

Fixes

• Extend try in agent loop to cover all job handling.
• Guard against undefined filestream timeout.
• Error if code artifact underlying job has been deleted.
• Use a unique name for the artifact created by verify.
• Use resume=allow when auto requeuing.
• Correct entrypoint path from disabled git repo subir.
• Override individual job resource_args.
• Fix import issue to support python 3.6.
• Fix override entrypoint when using sweeps on launch without a scheduler job.
• Fix resume reference when nil.
• Further speed up import time.
• Fix sample kubernetes agent manifest secret mount.
• Rm unused import.
• Watch to get kubernetes run statuses.
• Prohibit saving artifacts to a different project than their base artifact.
• Require existing artifacts to save to their source entity/project.
• Adjust system monitor start and stop functionality.
• Remove suspect characters when directory creation fails.
• Default log_code exclusion behavior now correctly handles wandb in the root path prefix.
• Disallow project queue creation.
• Catch all sweep set state errors.
• Create_job now works from jupyter notebook.
• Fix race condition for defer and update control.
• Improved handling and logging of tensor types.
• Launch cli command should exit with non-zero status if underlying launched run exits with non-zero status.
• Fix correctness for offline mode.
• Reports api - fix media_keys json path.
• Allow uint8 images to be logged as wandb.Image().
• Fall back to /tmp/username/.config/wandb in old settings.
• Use UpsertBucketRetryPolicy in all gql.UpsertBucket calls.
• Update report id validation and encoding.
• Add support for propagating messages from the internal process.

Cleanup

• Fix to new ruff rule E721 additions.

View the v0.15.10 changelog on GitHub.

​

v0.15.9

August 28, 2023

Enhancements

• Launch sweep schedulers to team queues from UI.
• Make vertex launcher more customizable.
• Default to noop builder if docker not installed.

Fixes

• Use built in entrypoint and args commands for sagemaker.
• Copy parent source project info to new draft artifact.
• Avoid error at end of run with bigints.
• Manually created image jobs can rerun correctly.

View the v0.15.9 changelog on GitHub.

​

v0.15.8

August 01, 2023

Enhancements

• Use mutation createRunFiles to get uploadUrls.
• Add create_run_queue to public API.
• Add hidden option to use orjson instead of json.
• Improve error message when building with noop builder.
• Create launch agent includes agent config if present.
• Check if job ingredients exist before making job.
• Gracefully handle Kubernetes 404 error.

Fixes

• Only creating new project if it doesn’t already exist.
• Support namespace in metadata key of resource args.
• Use "" instead of None for project kwarg when no project given.
• Add + to torch cpu regex + tests.
• Implement timeout for file_stream and add debug logs.
• Fix collection filtering when getting aliases.
• Replace dir_watcher settings with SettingsStatic.
• Set correct base for incremental artifacts.
• Drop https from azure registries to ensure compatibility with $ macro.
• Handle None description correctly.
• Don’t create k8s secret if it already exists.
• Drop S3 bucket versioning check.
• Speed up import time and fix pkg_resources DeprecationWarning.

View the v0.15.8 changelog on GitHub.

​

v0.15.7

July 25, 2023

Fixes

• Images not syncing until the end run (revert #5777).

View the v0.15.7 changelog on GitHub.

​

v0.15.6

July 24, 2023

Enhancements

• Add job link to wandb footer.
• Re-implement job requeueing, fixed cancel behavior.
• Manually create jobs from cli.
• Allow users to specify job name via the job_name setting.
• Add an simplified trace API to log prompt traces.
• Support .keras model format with WandbModelCheckpoint and TF 2.13.0 compatible.
• Initial support for migrating W&B runs and reports between instances.

Fixes

• Make LightGBM callback compatible with 4.0.0.
• Use default settings for project retrieval if available.

View the v0.15.6 changelog on GitHub.

​

v0.15.5

July 05, 2023

Enhancements

• Improve handling of docker image job names and tags.
• Support kaniko builds on AKS.
• Allow kaniko builds to run in other namespaces.
• Support access key for Azure references.
• Add information to failed run queue items, support warnings for run queue items.
• Allow direct configuration of registry uri for all registries.
• Enhance download URL fetching process with batch and retry logic.
• Add flag to skip missing S3 references in Artifact.download.
• Implement job requeueing when pod disconnects.
• Add setting to disable setproctitle.

Fixes

• Handle uri schemes in LogicalPath.
• Update object storage to include reference and prevent id reuse.
• Update click package version requirements.
• Improve lazy import to be thread-safe.
• Change typo in kaniko image name.
• Correct date parsing in SageMaker configuration.
• Make docker build non interactive to prevent region based questions.
• Update “cuda” base image path to “accelerator” base image path.
• Replace artifact name with placeholder to skip validation.
• Prevent jobs with large outputs from hanging on local-container.
• Ensure resume does not push sensitive info.
• Fix handling of references when downloading.
• Correct launch sweep author to personal username.
• Change artifact methods and attributes to private.

View the v0.15.5 changelog on GitHub.

​

v0.15.4

June 06, 2023

Enhancements

• Set job source in settings.
• Launch sweeps controlled by wandb run.
• Add autolog for Cohere python SDK.
• Support launching custom k8s objects.
• Conserve memory when hashing files.
• Add new_draft method to modify and log saved artifacts as new version.
• Don’t install frozen reqs if there is a reqs file.
• Don’t remove temp files from artifacts cache by default.
• Add source_entity and update sequenceName handling.
• Add ‘remove’ to Artifacts API.
• Optuna scheduler for sweeps on launch.
• Support notebook job creation.
• Enable launch macros for all runners.
• Add autologging for supported huggingface pipelines.
• Add usage metrics and table logging to OpenAI autologger.
• Add support for monitoring AMD GPU system metrics.
• Capture absolute GPU memory allocation.

Fixes

• Ensure wandb can be used in AWS lambda.
• Permit LogicalPath to strip trailing slashes.
• Exercise caution when creating ~/.config/wandb/settings file.
• Update custom chart query handling and add alternate constructor for table-based charts.
• Add s3 multipart uploading for artifact files.
• Handle incompatible artifact name strings.
• Docker runner always pull for image sourced jobs.
• Improve error handling for package installation.
• Custom k8s objects respect command/args overrides.
• Remove entity, project from valid properties and adjust name handling.
• Use env var for launch agent base url.
• Write to the cache defensively (catch OSError).
• Handle exception in finish_thread_id and fail run queue items.
• Add pull secrets for pre made images when registry is specified.
• Read kaniko pod sa name from env var.
• Misc gcp fixes.
• Support local environment and registry declaration.
• Support ssh git urls and submodules in agent.
• Update git repo handling for failure cases and rename to gitlib.
• Unify offline and online mode during init and fix multiprocess attach.
• Prevent errors by checking for wandb.run in Gym integration.
• Fix wandb tfevent sync issue.

View the v0.15.4 changelog on GitHub.

​

v0.15.3

May 17, 2023

Fixes

• Allow SDK to work if SA token can’t be read.
• Clean up the k8s token discovery logic in util.py::image_id_from_k8s.
• Update WandbTracer to work with new langchain version.
• Update summary for changed keys only.

View the v0.15.3 changelog on GitHub.

​

v0.15.2

May 05, 2023

Fixes

• Update WandbTracer for new langchain release by @parambharat @tssweeney in https://github.com/wandb/wandb/pull/5467.
• Fix error message in langchain wandb_tracer version check.

View the v0.15.2 changelog on GitHub.

​

v0.15.1

May 02, 2023

Enhancements

• Implement new Kubernetes runner config schema.
• Allow platform override for docker builder.
• Get full name of artifact for easier artifact retrieval.
• Make default root for artifacts download configurable.
• Add Azure storage handler in SDK.
• Add method to convert wandb.Table to pandas.DataFrame.
• Sweeps on launch command args passed as params.

Fixes

• Don’t assume keys in args and config refer to the same thing.
• Make ElasticContainerRegistry environment handle “ImageNotFoundException” gracefully.
• Disable kaniko builder retry.
• Refine error message for auth error.
• Kubernetes runner does not respect override args.
• Allow attr-dicts as sweeps configs.
• Checksum the read-only staging copy instead of the original file.
• Skip getting run info if run completes successfully or is from a different entity.
• Default to project “uncategorized” instead of “None” when fetching artifacts.
• Add enabled check to gym VideoRecorder.
• Fix handling of default project and entity.
• Update import_hook.py with latest changes in the wrapt repository.
• Fix support for local urls in k8s launch agent.
• Improve notebook environment detection and testing.
• Implement recursive isinstance check utility for the Settings object.
• Correctly parse edge cases in OpenMetrics filter definitions in System Monitor.
• Update debug logs to include SDK’s version.
• Filter AWS Trainium metrics by local rank if executed with torchrun.
• Inform users about WandbTracer incompatibility with LangChain > 0.0.153.

View the v0.15.1 changelog on GitHub.

​

v0.15.0

April 19, 2023

Enhancements

• Add support for LangChain media type.
• Add autolog for OpenAI’s python library by @dmitryduev @parambharat @kptkin @raubitsj in https://github.com/wandb/wandb/pull/5362.

Fixes

• Add function signature wrapper to the patched openai methods.
• Adjust OpenAI autolog public API to improve user experience by @dmitryduev @kptkin @raubitsj in https://github.com/wandb/wandb/pull/5381.

View the v0.15.0 changelog on GitHub.

​

v0.14.2

April 07, 2023

Fixes

• Fix wandb sync regression.

View the v0.14.2 changelog on GitHub.

​

v0.14.1

April 05, 2023

Enhancements

• Improve run.log_artifact() with default type and path references.
• Add opt-in support for async artifact upload.
• Update summary for changed keys only.
• Use a persistent session object for GraphQL requests.
• Allow setting of extra headers for the gql client.
• Allow filtering metrics based on OpenMetrics endpoints.

Fixes

• More informative message when failing to create staging artifact directory.
• Set default value for Kubernetes backoffLimit to 0.
• Remove default sorting when dumping config into a yaml file.
• Fix encoding for html types on windows.
• Clean up auto resume state when initializing a new run.
• Harden wandb.init() error handling for backend errors.
• Fix system monitor shutdown logic.
• Allow users to specify pinned versions in requirements.txt.
• Make wandb.log() handle empty string values properly.
• Raise exception when accessing methods and attributes of a finished run.

Cleanup

• Use ‘secrets’ module instead of custom random token generator.
• Move _manifest_json_from_proto to sender.py.

View the v0.14.1 changelog on GitHub.

​

v0.14.0

March 14, 2023

Enhancements

• Support cuda base image for launch runs.
• Warn users of which packages failed to install during build process.
• Add support for importing runs from MLFlow.
• Mark queued runs that fail to launch as FAILED.

Fixes

• Temporarily remove local api key validation.
• Launch agent gracefully removes thread when it has an exception.
• Give clear error message when cannot connect to Docker daemon.
• Launch support for EKS instance roles.
• Cleaner error messages when launch encounters docker errors and graceful fail.
• Hash docker images based on job version and dockerfile contents.
• Security(launch): warn when agent is started polling on a team queue.
• Add telemetry when syncing tfevents files.
• Fix regression preventing run stopping from working.
• Instruct user how to handle missing kubernetes import when using kubernetes runner or kaniko builder.
• Hide unsupported launch CLI options.
• Make launch image builder install Pytorch properly with dependencies on different hardware.

View the v0.14.0 changelog on GitHub.

​

v0.13.11

March 07, 2023

Enhancements

• Improve launch agent logging.
• Sweep run_cap now works for launch sweeps.
• Launch sweep jobs from image_uri.
• Add num_workers param to scheduler section in launch_config.
• Raise ArtifactNotLoggedError instead of ValueError.
• Launch agent uses thread pool to run jobs.
• Make runners and builders use Environment & Registry classes.
• Add OpenMetrics support for System Metrics.
• Add ability to filter system metrics consumed from OpenMetrics endpoints.
• Add support for gymnasium env monitoring, in addition to gym.
• Add max_scheduler key to launch agent config.
• Add an integration with ultralytics library for YOLOv8.

Fixes

• Clean up IPython’s widget deprecation warning.
• Add special Exceptions for the manager logic, when trying to connect to a gone service.
• Fix issue where global config directory had to be writable to use Api.
• Make error message during run initialization more actionable and fix uncaught exception.
• Add deepcopy dunder method to the Run class.
• Remove default to project always in sweep.
• Error out when trying to create a launch sweep without a job specified.
• Mkdir_exists_ok now (again) checks permission on existence.
• Only log the received job when launching something sourced from a job.
• Fix issue where queued runs sourced from images would vanish in URI.
• Add write permissions to copied artifacts.
• Improve queue argument parsing in sweep cli command.
• When in disable mode don’t spin up service.
• Fix support for docker images with user specified entrypoint in local container.
• API - ArtifactFiles no longer errors when accessing an item.
• Verify job exists before starting the sweeps scheduler.
• Handle system metrics requiring extra setup and teardown steps.
• Fix a typo in CONTRIBUTING.md.
• Correctly detect notebook name and fix code saving in Colab.
• Allow up to max_artifacts (fix off by 1 error).
• Exercise extra caution when starting asset monitoring threads.
• Fix bug where boto3 dependency crashes on import when downl….
• Verify num_workers cli arg is valid and default to 8 if not.
• Fix the file reference added to the verification artifact.
• Special handling for sweeps scheduler in agent.
• Only re-download or overwrite files when there are changes.
• Avoid introspection in offline mode.
• Topological ordering of wandb.Settings.
• Avoid lazy loading for tensorboard patching.

Cleanup

• Fix bugbear B028 add stacklevel.
• Move launch errors closer to the code.
• Move mailbox error closer to the code.
• Add unsupported error type.
• Add support for the ruff linter.
• Cosmetic changes for readability.
• Introduce environment and registry abstract classes.
• Fix unused union type in launch agent.
• Remove the artifact from the manifest.
• Enable typechecking for interface.artifacts and add type hints / casts.
• Type-annotate wandb_setup.py.
• Remove unused #noqa directives.
• Disable sentry tracking when testing.

View the v0.13.11 changelog on GitHub.

​

v0.13.10

February 07, 2023

Enhancements

• Reuse session for file upload requests.
• Expose aliases list endpoint for artifact collections.
• Include the username of the run’s author in the environment variables.
• Add support for local-container resource args.
• Add the ability to append to a run with wandb sync --append.
• Add an escape hatch (disable_job_creation) to disable automatic job creation.

Fixes

• Remove underscores from generated job name in kubernetes runner.
• Sweep command args can once again be int type.
• Ensure prepared artifacts have the latest alias.
• Catch FileNotFoundError and PermissionError during cache.cleanup().
• Fix order of python executable resolves.
• Fix console handling when forking and setting stdout==stderr.
• Fix issue where job artifacts are being logged without latest alias.
• Ensure job names do not exceed maximum allowable for artifacts.

Dev

• Ensure manifest version is verified.
• Add tests for custom SSL certs and disabling SSL.
• Fix nightly docker builds.
• Dont create universal py2/py3 package.
• Fix flake8-bugbear B028 and ignore B017.
• Fix gcloud sdk version requested in nightly tests.
• Remove unused parameters in StorageHandler.load_{path,file,reference}.
• Split unit tests to system tests and proper unit tests.
• Address fixture server move from port 9010 to 9015 in local-testcontainer.
• Add aliases to ac query response.
• Run regression suite nightly.
• Fix broken lightning test.
• Enable type checking for wandb_init.py.
• Deprecate defaulting to default queue in launch-agent command.
• Add unit test for kubernetes runner with annotations.
• Fix train_gpu_ddp test.
• Fix docker testimage to pull amd64 version.
• Fix codeowners after test restructure.
• Fix md5 test failures on Windows.
• Split out relay server so it can be shared with yea-wandb.
• Fix a flake8 complaint in a test.
• Fix several import tests.
• Don’t use symlinks for SSL test assets, because Windows.
• Add unit tests for filesync.Stats.
• Add async retry logic.
• Strengthen tests for ArtifactSaver, StepUpload.
• Agent logs full stack trace when catching exception.
• Swallow warning printed by neuron-ls.
• Pin pip and tox in development environments.

Cleanup

• Strengthen StepUpload tests; make exception-handling more thorough in upload/commit.
• Refactor Artifact query to fetch entity and project.
• Replace more communicate calls with deliver.
• Internally use Future to communicate success/failure of commit, not threading.Event.
• Use stdlib ThreadPoolExecutor in StepUpload instead of managing our own.

View the v0.13.10 changelog on GitHub.

​

v0.13.9

January 11, 2023

Fixes

• Exercise extra caution when checking if AWS Trainium is available in the system.
• Restore ‘util.generate_id’ for legacy / user code.
• Replace release with abandon when releasing mailbox handle during init.

View the v0.13.9 changelog on GitHub.

​

v0.13.8

January 10, 2023

Enhancements

• Keep uncommitted uploads in separate staging area.
• Improve file descriptor management.
• Default to using model-registry project for agent and launch_add.
• Add exist_ok=False to file.download().
• Auto create job artifacts from runs with required ingredients.
• Add generalized response injection pattern for tests.
• Replace multiprocessing.Queue’s with queue.Queue’s.
• Use transaction log to cap memory usage.
• Support system metrics for AWS Trainium.

Fixes

• Correct the type hint for wandb.run.
• Resume collecting system metrics on object restart.
• Fix env handling and node_selector handling.
• Fix Job.call() using the wrong keyword (queue vs queue_name) when calling launch_add.
• Sweeps schedulers handles multi word parameters.
• Allow spaces in requirements file, remove duplicate wandb bootstrap file.
• Correctly handle url-encoded local file references.
• Get digest directly instead of from the manifests’ manifest.
• Artifact.version should be the version index from the associated collection.
• Remove duplicate generate_id functions, replace shortuuid with secrets.
• Fix type check for jax.Array introduced in jax==0.4.1.
• Fix hang after failed wandb.init (add cancel).
• Allow users to provide path to custom executables.
• Fix TypeError when trying to slice a Paginator object.
• Add AttributeError to the list of handled exceptions when saving a keras model.
• Remove args from jobs.

Dev

• Update t2_fix_error_cond_feature_importances to install scikit-learn.
• Update base Docker images for nightly testing.
• Change sklearn to scikit-learn in functional sacred test.
• Add error check for --build when resource=local-process.
• Update scheduler and agent resource handling to allow DRC override.
• Require sdk-team review for adding or removing high-level….
• Remove requirement to make target project match queue.
• Enhance nightly cloud testing process.
• Update pull request template.
• Return updated runSpec after pushToRunQueue query.
• Fix for run spec handling in sdk.
• Remove test dependency on old fastparquet package.
• Fix dtype np.float (does not exist), set to python float.
• Correct ‘exclude’ to ‘ignore-paths’ in .pylintrc.
• Use pytest tmp_path so we can inspect failures.
• Reset build command after building.
• Rerun flaking tests in CI with pytest-rerunfailures.
• Remove dead code from filesync logic.
• Remove unused fields from a filesync message.
• Refactor retry logic to use globals instead of dependency-injecting them.
• Add unit tests for filesync.StepUpload.
• Add tests for Api.upload_file_retry.
• Remove fallback resource when not specified for a queue.
• Improve storage handler test coverage.
• Fix import tests.
• Make MetricsMonitor less verbose on errors.
• Address fixture server move from port 9003 to 9010 in local-testcontainer.
• Vendor promise==2.3.0 to unequivocally rm six dependency.
• Allow setting artifact cache dir in wandb.init(…).
• Temporary lower network buffer for testing.
• Add telemetry if the user running in pex environment.
• Add more flow control telemetry.
• Add settings and debug for service startup issues (wait_for_ports).
• Fix AWS Trainium test.
• Fix status checker thread issue when user process exits without finish().
• Add telemetry for service disabled usage.

Cleanup

• Use the same syntax whenever raising exceptions.
• Combine _safe_mkdirs with mkdir_exist_ok.
• Use a pytest fixture for the artifact cache.
• Use ArtifactEntry directly instead of subclassing.
• Consolidate hash utilities into lib.hashutil.
• Format public file with proper formatting.
• Install tox into proper env in dev env setup tool.
• Clean up the init and run logic.

View the v0.13.8 changelog on GitHub.

​

v0.13.7

December 14, 2022

Fixes

• Revert Circular reference detected change to resolve Object of type Tensor is not JSON serializable.

View the v0.13.7 changelog on GitHub.

​

v0.13.6

December 06, 2022

Enhancements

• Add Sweep.expected_run_count to public Api.
• Support volume mounts and security contexts in kubernetes runner.
• Add a new --build flag for building and then pushing the image to a queue.
• Add ability to log learning rate using WandbMetricsLogger.
• Improve Report API in preparation for GA.

Fixes

• Add filter for artifact_version to only retrieve committed artifacts.
• Deflake wandb verify.
• Fix the type of the override args passed through to a LaunchProject from a Job.
• Remove extra colon from log prefix.
• Add support for service running in a pex based environment.
• Fix probing static IPU info.
• Change artifactSequence to artifactCollection in public GQL requests.
• Fix TF compatibility issues with WandbModelCheckpoint.
• Make Keras WandbCallback compatible with TF version >= 2.11.0.
• Update gym integration to match last version.
• Harden internal thread management in SystemMetrics.

Dev

• Improved git fetch time by specifying a refspec and depth=1.
• Fix linguist rule to ignore grpc generated files.
• New shard for launch tests.
• Upgrade Node 12 based GitHub Actions.
• Skip flaky artifact_metadata_save test.
• Replace sleeps with flush when waiting on a file to write.
• Use tmp_path fixture instead of writing local files during tests.
• Fix broken queue test.
• skip instead of xfail for test test_artifact_metadata_save.
• Add many tests for InternalApi.upload_file.
• Add artifact Sequence fallback for older servers.
• Make protobuf version requirements more granular.

Cleanup

• When committing artifacts, don’t retry 409 Conflict errors.
• Add programmatic alias addition/removal from SDK on artifacts.
• Remove wandb.sklearn.plot_decision_boundaries that contains dead logic.
• Adds an option to force pull the latest version of a test dev-container image.
• Noop builder.
• Remove unused attribute.
• Update mypy to 0.991.
• Add more robust uri parsing.
• Turn on linting for internal_api.py.
• Remove dependency on six by modifying vendored libs.

View the v0.13.6 changelog on GitHub.

​

v0.13.5

November 03, 2022

Enhancements

• Add an option to upload image references.

Fixes

• Generate more readable image names.
• Use hash(etag+url) instead of just etag, as key, in artifacts cache.
• Wait for artifact to commit before telling the user it’s ready when using wandb artifact put.
• Prefix vendor watchdog library.
• Fix Circular reference detected error, when updating metadata with numpy array longer than 32 elements.
• Add a random string to run_id on SageMaker not to break DDP mode.

Dev

• Make sure we dont shutdown test cluster before grabbing results.
• Add standalone artifact test to nightly cpu suite.
• Rename default branch to main.
• Update mypy extension for protobuf type checking.
• Update codeql-analysis.yml branch name.
• Move functional import tests to nightly and expand python version coverage.
• Add Slack notification for failed nightly import tests.
• Fix broken CLI tests that attempt uploading non-existent artifacts.

Cleanup

• Job creation through use_artifact instead of log_artifact.
• Add a GH action to automate parts of the release process.
• 3D Point Clouds now viewable in UI in all situations.
• Git URLs were failing if fsmonitor is enabled.
• Ignore new proto generated file directories.
• Fix a bug preventing Run Queue deletion in the SDK.
• Add support for pushToRunQueueByName mutation.
• Refactor system metrics monitoring and probing.
• Fix gitattribute for protobuf generated files.

View the v0.13.5 changelog on GitHub.

​

v0.13.4

October 05, 2022

Enhancements

• Show entity and project in k8s job names.
• Add environment variable sweep command macro.
• Add from_* constructors and scene camera and bounding box confidence scores to Object3D data type.
• Add simple progress indicator for artifact downloads.
• Add WandbMetricsLogger callback - a Keras dedicated metrics logger callback.
• Add WandbModelCheckpoint callback - a Keras model checkpointing callback.
• Add WandbEvalCallback callback - a Keras callback for logging model predictions as W&B tables.

Fixes

• Cast agent’s config max_jobs attribute to integer.
• Correct the displayed path to the debug-cli.log (debug log).
• Catch retry-able request timeout when uploading artifacts to AWS.
• Improve user feedback for long running calls: summary, finish.
• Fix RuntimeError when using keras.WandbCallback with tf.MirroredStrategy.

Dev

• Add code analysis/scanning with codeql.
• Validate PR titles to ensure compliance with Conventional Commits guidelines.
• Harden launch by pining the build versions of kaniko and launch-agent-dev.
• Add telemetry for the mmengine package.
• Add the build type to our conventional commits setup.
• Add tensorflow_datasets requirement to imports12 shard.
• Fix sb3 test by pinning upstream requirement.
• Make the SDK compatible with protobuf v4.
• Fix flake8 output coloring.
• Fix artifact reference test asset directory.

Cleanup

• Fix type hint for filters argument in public_api.runs.
• Improve type annotations around artifact-file-creation.
• Improve type annotations and VSCode config for public API.
• Make type annotations more easily navigable in VSCode.
• Introduce str NewTypes and use them for various Artifact fields.
• Add type annotations to get better IDE hints for boto3 usage.

View the v0.13.4 changelog on GitHub.

​

v0.13.3

September 08, 2022

Enhancement

• Adds raytune examples / tests.
• Refactors pytest unit tests to run against real wandb server.
• Adds Launch kubernetes support of taints and tolerations.
• Adds Sweeps on Launch on Kubernetes.
• Adds parallelism to functional testing.
• Upgrades mypy to version 0.971.
• Adds Mailbox async internal process communication.
• Implements searching launch job in sweep config.
• Improves performance when sending large messages.
• Vendors the latest nvidia-ml-py-11.515.48.
• Improves performance by increase recv size on service socket.
• Adds isort support with black profile.
• Implements pushing test-results to CircleCI for nightly tests.
• Adds debug mode for pytest unit tests.
• Adds support for arguments in Launch Jobs.
• Adds FetchRunQueueItemById query.
• Adds telemetry for keras-cv.
• Adds sentry session tracking.
• Adds the ability to log artifact while linking to registered model.

Cleanup

• Breaks gradient and parameters hooks.
• Adds explicit error message for double uri/docker-image.
• Tests that the wandb_init fixture args are in sync with wandb.init().
• Upgrades the GKE cluster used for nightly tests to n1-standard-8.
• Moves service teardown to the end of tests.
• Reduce the pytest job parallelism from 10 to 6.
• Removes service user doc.
• Move _timestamp logic to the internal process.
• Adds Launch gitversion error message.
• Updates KFP machine VM image in CircleCI.
• Upgrades sweeps to latest version.
• Implements Sweep scheduler cleanup and better tests.
• Adds a requirement for the sdk-team to approve API changes.
• Adds additional time for artifact commit.
• Implements tox configuration with dynamic resolution.
• Removes buildx version pin for nightly builds.
• Moves Launch run configs from entrypoint into params.
• Removes Slack orb usage from Win job on CircleCI.
• Adds heartbeat parsing for Launch run args using legacy agent.
• Add better error handling when tearing down service.
• Cleans up Launch job creation pipeline.
• Adds detail to error message when uploading an artifact with the wrong type.
• Adds optional timeout parameter to artifacts wait().
• Sanitizes numpy generics in keys.
• Removes reassignment of run function in public api.
• Makes pulling sweeps optional when using public api to query for runs.
• Updates ref docs for wandb.init to give more info on special characters.

Bug Fix

• Fixes Sweeps on Launch Jobs requirement.
• Fixes Artifact metadata JSON-encoding to accept more types.
• Adjusts root_dir setting processing logic.
• Prevents run.log() from mutating passed in arguments.
• Fixes 05-batch5.py test.
• Allows users to control the run_id through the launch spec.
• Fixes accidental overwrite in config.yml.
• Ensures propagating overridden base_url when initializing public API.
• Fixes Sweeps on Launch CLI launch config, relpath.
• Fixes broken Launch apikey error message.
• Marks flakey sweeps test xfail.
• Fixes Launch gitversion error message.
• Fixes yea-wandb dev release -> release.
• Cleans up outstanding issues after the client->wandb rename.
• Fixes test precision recall.
• Fixes functional sklearn test.
• Fixes hang caused by keyboard interrupt on windows.
• Fixes default test container tag.
• Fixes summary handling in conftest.py.
• Fixes some small typos in cli output.
• Fixes issue triggered by colab update by using default file and catching exceptions.
• Fixes mailbox locking issue.
• Fixes variable inclusion in log string.
• Corrects wandb.Artifacts.artifact.version attribute.
• Fixes piping of docker args by Launch Agent.
• Fixes RecursionError when printing public API User object without email fetched.
• Fixes deserialization of numeric column names.

View the v0.13.3 changelog on GitHub.

​

v0.13.2

August 22, 2022

Bug Fix

• Fix issue triggered by colab update by using default file and catching exceptions.

View the v0.13.2 changelog on GitHub.

​

v0.13.1

August 05, 2022

Bug Fix

• Prevents run.log() from mutating passed in arguments.

View the v0.13.1 changelog on GitHub.

​

v0.13.0

August 04, 2022

Enhancement

• Turns service on by default.
• Adds support logic for handling server provided messages.
• Allows runs to produce jobs on finish.
• Adds Job, QueuedRun and job handling in launch.
• Supports in launch agent of instance roles in ec2 and eks.
• Adds default behavior to the Keras Callback: always save model checkpoints as artifacts.
• Sanitizes the artifact name in the KerasCallback for model artifact saving.
• Improves console logging by moving emulator to the service process.
• Fixes data corruption issue when logging large sizes of data.
• Adds the state to the Sweep repr in the Public API.
• Adds an option to specify different root dir for git using settings or environment variables.
• Adds an option to pass remote url and commit hash as arguments to settings or as environment variables.
• Improves time resolution for tracked metrics and for system metrics.
• Defaults to project name from the sweep config when project is not specified in the wandb.sweep() call.
• Adds support to use namespace set user by the the launch agent.
• Adds telemetry to track when a run might be overwritten.
• Adds a tool to export wandb’s history into sqlite.
• Replaces some Mapping[str, ...] types with NamedTuples.
• Adds import hook for run telemetry.
• Implements profiling support for IPUs.

Bug Fix

• Fixes sweep agent with service.
• Fixes an empty type equals invalid type and how artifact dictionaries are handled.
• Fixes wandb.Config object to support default values when getting an attribute.
• Removes default config from jobs.
• Fixes an issue where patch is None.
• Fixes requirements.txt parsing in nightly SDK installation checks.
• Fixes 409 Conflict handling when GraphQL requests timeout.
• Fixes service teardown handling if user process has been terminated.
• Adds storage_path and fixed artifact.files.
• Fixes performance issue syncing runs with a large number of media files.

Cleanup

• Adds an escape hatch logic to disable service.
• Annotates wandb/docker and reverts change in the docker fixture.
• Fixes GFLOPS to GFLOPs in the Keras WandbCallback.
• Adds type-annotate for file_stream.py.
• Renames repository from client to wandb.
• Updates documentation: adding --report_to wandb for HuggingFace Trainer.
• Makes aliases optional in link_artifact.
• Renames wandb local to wandb server.
• Updates README badges.

View the v0.13.0 changelog on GitHub.

​

v0.12.21

July 05, 2022

Enhancement

• Fixes config not showing up until the run finish.
• Adds new types to the TypeRegistry to handling artifact objects in jobs and run configs.
• Adds new query to the the internal api getting the state of the run.
• Replaces unsafe yaml loaders with yaml.safe_load.
• Improves testing tooling by allowing to specify shards in manual testing.
• Fixes ROC and PR curves in the sklearn integration by stratifying sampling.
• Fixes input box in notebooks exceeding cell space.
• Allows string to be passed as alias to link_model.
• Adds Support for FLOPS Calculation in keras’s WandbCallback.
• Extends python report editing.

Bug Fix

• Fixes stats logger so it can find all the correct GPUs in child processes.
• Fixes regression in s3 reference upload for folders.
• Fixes artifact commit logic to handle collision in the backend.
• Checks for None response in the retry logic (safety check).
• Adds sweeps on top of launch (currently in MVP).
• Renames functional tests dir and files.

Cleanup

• Fixes conditions order of _to_dict helper.
• Fixes changelog broken link to PR 3709.
• Fixes public api query (QueuedJob Api ).
• Renames local runners to local-container and local-process.
• Adds type annotations to files in the wandb/filesync directory.
• Re-organizes all the testing directories to have common root dir.
• Fixes testing configuration and add bigger machine on CircleCi.
• Fixes typo in the wandb-service-user readme file.
• Fixes broken artifact test for regression.
• Removes unused files (relating to py27) and empty submodules declaration.
• Adds extra for model reg dependency on cloudpickle.
• Replaces deprecated threading aliases.
• Updates the sdk readme to the renamed (local -> server) commands.

View the v0.12.21 changelog on GitHub.

​

v0.12.20

June 29, 2022

Bug Fix

• Retry commit_artifact on conflict-error.

View the v0.12.20 changelog on GitHub.

​

v0.12.19

June 22, 2022

Bug Fix

• Fix regression in s3 reference upload for folders.

View the v0.12.19 changelog on GitHub.

​

v0.12.18

June 09, 2022

Enhancement

• Launch: BareRunner based on LocalRunner.
• Add ability to specify api key to public api.
• Add support in artifacts for files with unicode on windows.
• Added telemetry for new packages.
• Improve API key management.
• Add information about wandb server during login.

Bug Fix

• Natively support timestamps in Python Table Types.
• Add support for magic with service.
• Add unit tests for DirWatcher and supporting classes.
• Improve DirWatcher.update_policy O(1) instead of O(num files uploaded).
• Add argument to control what to log in SB3 callback.
• Improve parameter naming in sb3 integration.
• Adjust the requirements for the dev environment setup on an M1 Mac.
• Launch: Fix NVIDIA base image Linux keys.
• Fix launch run queue handling from config file.
• Fix issue where tfevents were not always consumed.
• [Snyk] Fix for 8 vulnerabilities.
• Fix s3 storage handler to upload folders when key names collide.
• Correctly load timestamps from tables in artifacts.
• Require protobuf<44.
• Make Containers created through launch re-runnable as container jobs.
• Fix tensorboard integration skipping steps at finish().
• Rename wandb local to wandb server.
• Fix busted docker inspect command.
• Add dedicated sentry wandb.
• Image Type should gracefully handle older type params.

Cleanup

• Inline FileEventHandler.synced into the only method where it’s used.
• Use passed size argument to make PolicyLive.min_wait_for_size a classmethod.
• Make FileEventHandler an ABC, remove some “default” method impls which were only used once.
• Remove unused field from DirWatcher.
• Make sweeps an extra instead of vendoring.
• Add nightly CI testing.
• Improve keras and data type Reference Docs.
• Update pytorch version requirements in dev environments.
• Clean up CircleCI config.
• Add py310 testing in CI.
• Ditch dateutil from the requirements.
• Add deprecated string to Table.add_row.

View the v0.12.18 changelog on GitHub.

​

v0.12.17

May 26, 2022

Bug Fix

• Update requirements to fix incompatibility with protobuf >= 4.

View the v0.12.17 changelog on GitHub.

​

v0.12.16

May 03, 2022

Enhancement

• Improve W&B footer by aligning summary/history in notebook env.
• Enable experimental history step logging in artifacts.
• Add args_no_boolean_flags macro to sweep configuration.
• Add logging support for jax.bfloat.bfloat16.
• Raise exception when Table size exceeds limit.
• Add kaniko k8s builder for wandb launch.
• Add wandb.init() timeout setting.
• Do not assume executable for given entrypoints with wandb launch.
• Jupyter environments no longer collect command arguments.
• Add support for TensorFlow/Keras SavedModel format.

Bug Fix

• Support version IDs in artifact refs, fix s3/gcs references in Windows.
• Fix support for multiple finish for single run using wandb-service.
• Fix duplicate backtrace when using wandb-service.
• Fix wrong entity displayed in login message.
• Fix hang when wandb.init is interrupted mid setup using wandb-service.
• Fix handling keyboard interrupt to avoid hangs with wandb-service enabled.
• Fix console logging with very long print out when using wandb-service.
• Fix broken artifact string in launch init config.

Cleanup

• Fix typo in wandb.log() docstring.
• Cleanup custom chart code and add type annotations to plot functions.
• Improve wandb.init(settings=) to handle Settings object similarly to dict parameter.
• Add documentation note about api.viewer in api.user() and api.users().
• Be explicit about us being py3+ only in setup.py.
• Add type annotations to DirWatcher.
• Improve wandb.log() docstring to use the correct argument name.

View the v0.12.16 changelog on GitHub.

​

v0.12.15

April 21, 2022

Enhancement

• Optimize wandb.Image logging when linked to an artifact.

View the v0.12.15 changelog on GitHub.

​

v0.12.14

April 08, 2022

Bug Fix

• Fix regression: disable saving history step in artifacts.

View the v0.12.14 changelog on GitHub.

​

v0.12.13

April 07, 2022

Bug Fix

• Revert strictened api_key validation.

View the v0.12.13 changelog on GitHub.

​

v0.12.12

April 05, 2022

Enhancement

• Allow run objects to be passed to other processes when using wandb-service.
• Add create user to public api.
• Support logging from multiple processes with wandb-service.
• Add gpus flag for local launch runner with cuda.
• Improve Launch deployable agent.
• Add Launch kubernetes integration.
• KFP: Add wandb visualization helper.
• KFP: Link back to Kubeflow UI.
• Add boolean flag arg macro.

Bug Fix

• Improve host / WANDB_BASE_URL validation.
• Fix/insecure tempfile.
• Fix excess warning span if requested WANDB_DIR/root_dir is not writable.
• Fix line_series to plot array of strings.
• Properly handle command line args with service.
• Improve api_key validation.
• Fix multiple performance issues caused by not using defaultdict.
• Enable inf max jobs on launch agent.
• Fix colab command to work with launch.
• Fix typo in Config docstring.
• Make code saving not a policy, keep previous custom logic.
• Fix logging sequence images with service.
• Add username to debug-cli log file to prevent conflicts of multiple users.
• Fix python sweep agent for users of wandb service / pytorch-lightning.
• Remove unnecessary launch reqs checks.
• Workaround for MoviePy’s Unclosed Writer.
• Improve handling of Run objects when service is not enabled.

View the v0.12.12 changelog on GitHub.

​

v0.12.11

March 01, 2022

Enhancement

• Add captions to Molecules.
• Add CatBoost Integration.
• Launch: AWS Sagemaker integration.
• Launch: Remove repo2docker and add gpu support.
• Adds Timestamp inference from Python for Weave.
• Launch GCP vertex integration.
• Use Artifacts when put into run config. Accept a string to represent an artifact in the run config.
• Improve xgboost wandb_callback (#2929).
• Add initial kubeflow pipeline support.

Bug Fix

• Fix logging of images with special characters in the key.
• Fix azure blob upload retry logic.
• Fix program field for scripts run as a python module.
• Fix issue where sync_tensorboard could die on large histograms.
• Fix wandb service performance issue during run shutdown.
• Fix vendoring of gql and graphql.
• Flush log data without finish with service.
• Fix wandb service hang when the service crashes.
• Fix issue logging images with ”/” on Windows.
• Add image filenames to images/separated media.
• Add setproctitle to requirements.txt.
• Fix issue where sagemaker run ids break run queues.
• Fix encoding exception when using %%capture magic.

View the v0.12.11 changelog on GitHub.

​

v0.12.10

February 01, 2022

Enhancement

• Improve validation when creating Tables with invalid columns from dataframes.
• Enable digest deduplication for use_artifact() calls.
• Initial prototype of azure blob upload support.

Bug Fix

• Fix wandb launch using python dev versions.
• Fix loading table saved with mixed types.
• Fix ResourceWarning when calling wandb.log.
• Fix missing cursor in ProjectArtifactCollections.
• Fix windows table logging classes issue.
• Gracefully handle string labels in wandb.sklearn.plot.classifier.calibration_curve.
• Do not display login warning when calling wandb.sweep().

Cleanup

• Drop python2 backport deps (enum34, subprocess32, configparser).
• Settings refactor.

View the v0.12.10 changelog on GitHub.

​

v0.12.9

December 16, 2021

Bug Fix

• Fix regression in upload_file() exception handler.

View the v0.12.9 changelog on GitHub.

​

v0.12.8

December 16, 2021

Enhancement

• Update contributing guide and dev env setup tool.
• Improve wandb_callback for LightGBM (#2945).

Bug Fix

• Reduce GPU memory usage when generating histogram of model weights.
• Support mixed classes in bounding box and image mask annotation layers.
• Add max-jobs and launch async args.
• Support lists of Summary objects encoded as strings to wandb.tensorboard.log.
• Fix handling of 0 dim np arrays.
• Fix handling of empty default config file.
• Add service backend using sockets (support fork).
• Send git port along with url when sending git repo.
• Add support raw ip addresses for launch.
• Tables no longer serialize and hide 1d NDArrays.
• Fix artifact file uploads to S3 stores.
• Send uploaded file list on file stream heartbeats.
• Add support for keras experimental layers.
• Fix from wandb import magic to not require tensorflow.
• Fix launch permission error.

View the v0.12.8 changelog on GitHub.

​

v0.12.7

November 18, 2021

Bug Fix

• Fix issue where console log streaming was causing excessive network traffic.
• Metaflow: Make optional dependencies actually optional.
• Fix docstrings for wandb.watch and ValidationDataLogger.
• Prevent launch agent from sending runs to a different project or entity.
• Fix logging pr_curves through tensorboard.
• Prevent TPU monitoring from reporting invalid metrics when not available.
• Make import order dependencies for WandbCallback more robust.
• Fix a bug in feature importance plotting to handle matrices of different shapes.
• Fix base url handling to allow trailing /.
• Prevent wandb.agent() from sending too many heartbeats impacting rate limits.
• Redact sensitive information from debug logs.

Enhancement

• Add wandb.Molecule support for rdkit supported formats.
• Add module-level docstrings for reference doc modules.
• Store launch metadata in file.
• Add Project.sweeps() public API call to view all sweeps in a project.
• Ensures API key prompt remains captive when user enters nothing.
• Refactors wandb.sklearn into submodules.
• Support code artifacts in wandb launch.
• Improve launch agent (async, stop, heartbeat updates).
• Improve usage and error messages for anonymous mode.
• Add example on how to find runs with wandb.Api().runs(…) matching a regex.

View the v0.12.7 changelog on GitHub.

​

v0.12.6

October 27, 2021

Bug Fix

• Fix sklearn plot_calibration_curve() issue breaking the provided model.
• Fix CondaEnvExportError by redirecting stderr.
• Fix use_artifact() when specifying an artifact from a different project.

Enhancement

• Add metric names to pr curve charts in tensorboard.

View the v0.12.6 changelog on GitHub.

​

v0.12.5

October 19, 2021

Bug Fix

• Report errors for invalid characters in logged media keys on windows.
• Handle errors when probing for TPUs in unsupported environments.
• Fixed bug where %%wandb followed by wandb.init() does not display run links.
• Fixed api.runs() to correctly return all runs for the current entity/project.

Enhancement

• Add wandb.require(experiment="service") to improve multiprocessing support.
• Add support for swappable artifacts in launch context.
• Add wandb.login(timeout=) support for jupyter environments.
• Add ability to disable git ref saving with WANDB_DISABLE_GIT.
• Support newer versions of pytest-mock and PyYAML.
• Add ability to delete artifacts with aliases: artifact.delete(delete_aliases=True).
• Add unwatch() method to the Run object.

View the v0.12.5 changelog on GitHub.

​

v0.12.4

October 05, 2021

Bug Fix

• Fix regression introduced in 0.12.2 causing network access when WANDB_MODE=offline.

View the v0.12.4 changelog on GitHub.

​

v0.12.3

September 30, 2021

Bug Fix

• Fixes the grid search stopping condition in the local controller.

Enhancement

• New jupyter magic for displaying runs, sweeps, and projects %wandb path/to/run -h 1024.
• We no longer display run iframe by default in jupyter, add %%wandb to a cell to display a run.
• Makes api key prompting retry indefinitely on malformed input.
• Invite users to teams via the api api.team("team_name").invite("username_or_email").
• Remove users from a team via the api api.team("team_name").members[0].delete().
• Create service accounts via the api api.team("team_name").create_service_account("Description").
• Manage api keys via the api api.user("username_or_email").generate_api_key().
• Add pytorch profiling trace support with wandb.profiler.torch_trace_handler().

View the v0.12.3 changelog on GitHub.

​

v0.12.2

September 15, 2021

Bug Fix

• Fix tensorboard_sync to handle ephemeral Sagemaker tfevents files.
• Fix Reports query from the public api (broken pagination and report path).
• Fix wandb.login() when relogin is specified (only force login once).

Enhancement

• Clean up footer output of summary and history metrics.
• Clean up error message from wandb sweep --update.
• Add warning for wandb local users to update their docker.
• Add optional argument log_learning_curve to wandb.sklearn.plot_classifier().
• Restore frozen pip package versions when using wandb launch.
• Add support for jupyter notebooks in launch.
• Add wandb.login() timeout option.

View the v0.12.2 changelog on GitHub.

​

v0.12.1

August 26, 2021

Bug Fix

• Fix tensorflow/keras 2.6 not logging validation examples.
• Fix metrics logged through tensorboard not supporting time on x-axis.
• Fix WANDB_IGNORE_GLOBS environment variable handling.
• Fix handling when sys.stdout is configured to a custom logger.
• Fix sklearn feature importance plots not matching feature names properly.
• Fix an issue where colab urls were not being captured.
• Save program commandline if run executable was outside cwd.

Enhancement

• Add Prodigy integration to upload annotated datasets to W&B Tables.
• Add initial Metaflow support.
• Add experimental wandb launch support.
• Add warnings that public API requests are timing out and allow override.
• Improve error handling in local controller sweeps engine.

View the v0.12.1 changelog on GitHub.

​

v0.12.0

August 10, 2021

No Longer Supported

• Remove Python 3.5 support.

Bug Fix

• Fix issue that could cause artifact uploads to fail if artifact files are being modified.
• Fix issue where wandb.restore() wouldn’t work with runs from a sweep.

Enhancement

• Improve run execution time calculation.

View the v0.12.0 changelog on GitHub.

​

v0.11.2

August 02, 2021

Bug Fix

• Restore vendored graphql-core library because of network regression.

View the v0.11.2 changelog on GitHub.

​

v0.11.1

July 29, 2021

Deprecated

• Python 3.5 will not be supported as of wandb==0.12.0.

Bug Fix

• Reduce Memory Footprint of Images In Tables.
• Added a dependency on graphql-core>=2.3.0.
• Removed urllib3 pin to avoid conflicts, if you see urllib3 related errors run pip install --upgrade urllib3.
• Improved Public API HTTP error messages.
• Set run.dir to the generated directory name in disabled mode.

Enhancement

• Adds support for native Jax array logging.
• Tables now support Molecule data type.
• Improve Stable-Baselines3 API by auto log model’s name and always upload models at the end of training.
• Implements the sweep local controller using wandb/sweeps.

View the v0.11.1 changelog on GitHub.

​

v0.11.0

July 15, 2021

No Longer Supported

• Remove Python 2.7 support.

Bug Fix

• Fix issue where wandb.watch() broke model saving in pytorch.
• Fix issue where uniform sweep parameters were parsed as int_uniform.
• Fix issue where file_stream thread was killed on 4xx errors.

Enhancement

• Improve performance of artifact logging by making it non-blocking.
• Add wandb integration for Stable-Baselines3.
• Improve keras callback validation logging inference logic.
• Expose sweep state via the public API.
• Improve performance of sweep run fetches via the API.

View the v0.11.0 changelog on GitHub.

​

v0.10.33

June 28, 2021

Bug Fix

• Fix issue where wandb restore 404ed if the run did not have a diff.patch file.
• Fix issue where wandb.log raised an Exception after trying to log a pandas dataframe.
• Fix issue where runs could be marked finished before files were finished uploading.

Enhancement

• Disable reloading of run metadata (such as command) in resumed runs.
• Allow logging of pandas dataframes by automatically converting them to W&B tables.
• Fix up log_code() exclude fn to handle .wandb dir.
• Improve handling of PyTorch model topology.
• Increase config debounce interval to 30s to reduce load on WB/backend.
• Improve reliability of CLI in generating sweeps with names, programs, and settings.

View the v0.10.33 changelog on GitHub.

​

v0.10.32

June 10, 2021

Bug Fix

• Make log_artifact() more resilient to network errors.
• Removed Duplicate Artifact Dependencies.
• Workaround urlib3 issue on windows.
• Fix regression where ipython was hanging.
• Allow logging of numpy high precision floating point values.
• Reduce liklyhood of collisions for file backed media or artifact objects.
• Fix wandb.watch() regression when logging pytorch graphs.

Enhancement

• Add support for logging joined and partitioned table.
• Handle schema validation warnings for sweep configs.
• Improve wandb sync to handle errors.
• Add ability to label scripts and repositories who use wandb.

View the v0.10.32 changelog on GitHub.

​

v0.10.31

May 27, 2021

Bug Fix

• Wandb.login() did not properly persist the host parameter.
• Fix issue where step information was not synced properly when syncing tensorboard directories.
• Fix some unicode issues with python2.7.
• Fixed bug in plot_calibration_curve for ComplementNB.
• Fall back to not using SendFile on some linux systems.
• Fix console issues where lines were truncated.
• Fix console issues where console logging could block.

Enhancement

• Add support for preemptible sweeps.
• Add command line for sweep control.
• Add support to load artifact collection properties.

View the v0.10.31 changelog on GitHub.

​

v0.10.30

May 07, 2021

Bug Fix

• Found and fixed the remaining issues causing runs to be marked crashed during outages.
• Improved performance for users of define_metric, pytorch-lightning, and aggressive config saving.
• Fix issue when trying to log a cuda tensor to config or summary.
• Remove dependency on torch backward_hooks to compute graph.
• Fix an issue preventing the ability to resume runs on sagemaker.
• Fix issues preventing pdb from working reliably with wandb.
• Fix deprecation warning in vendored library (user submission).
• Fix logging behavior where the library was accidently outputting logs to the console.
• Fix disabled mode to not create wandb dir and log files.
• Renamed types to prep for Tables launch.

Enhancement

• Allow renaming groups with public api.

View the v0.10.30 changelog on GitHub.

​

v0.10.29

May 03, 2021

Bug Fix

• Fix more network handling issues causing runs to be marked crashed (wandb sync to recover).
• Improve logging and exception handling to improve reporting and logging of crashed processes.

View the v0.10.29 changelog on GitHub.

​

v0.10.28

April 28, 2021

Bug Fix

• Fix network handling issue causing runs to be marked crashed (wandb sync to recover).
• Use register_full_backward_hook to support models with Dict outputs.
• Allow periods in table columns.
• Fix artifact cache collisions when using forked processes.
• Fix issue where custom charts do not display properly with pytorch-lightning.

Enhancement

• Add experimental incremental artifact support.
• Improve warnings when logging is being rate limited.

View the v0.10.28 changelog on GitHub.

​

v0.10.27

April 19, 2021

Bug Fix

• Fix tensorboard_sync condition where metrics at end of short run are dropped.
• Fix wandb sync when tensorboard files are detected.
• Fix api key prompt in databricks notebook.

Enhancement

• Integrate DSViz into Keras WandbCallback.
• Add support for conda dependencies (user submit).

View the v0.10.27 changelog on GitHub.

​

v0.10.26

April 13, 2021

Bug Fix

• Fix network handling issue where syncing stopped (use wandb sync to recover).
• Fix auth problem when using sagemaker and hugginface integrations together.
• Fix handling of NaN values in tables with non floats.
• Lazy load API object to prevent unnecessary file access on module load.

Enhancement

• Improve error messages when using public api history accessors.

View the v0.10.26 changelog on GitHub.

​

v0.10.25

April 05, 2021

Bug Fix

• Fix possible artifact cache race when using parallel artifact reads.
• Fix artifact reference when checksum=False.

Enhancement

• Release run.define_metric() to simplify custom x-axis and more.
• Add column operators add_column, get_column, get_index to wandb.Table().

View the v0.10.25 changelog on GitHub.

​

v0.10.24

March 30, 2021

Bug Fix

• Significant fixes to stdout/stderr console logging.
• Prevent excessive network when saving files with policy=live.
• Fix errors when trying to send large updates (most common with wandb sync).

Enhancement

• Automatically generate run_table artifact for logged tables.
• Add bracket notation to artifacts.
• Improve URL validation when specifying server url to wandb login.

View the v0.10.24 changelog on GitHub.

​

v0.10.23

March 22, 2021

Bug Fix

• Fix logged artifacts to be accessible after wait().
• Fix spell.run integration.
• Performance fix syncing console logs with carriage returns.
• Fix confusion matrix with class names and unlabeled data.

Enhancement

• Add the ability to save artifacts without creating a run.
• Add Foreign Table References to wandb.Table.
• Allow the same runtime object to be logged to multiple artifacts.
• Add experimental run._define_metric() support.
• Warn and ignore unsupported multiprocess wandb.log() calls.

View the v0.10.23 changelog on GitHub.

​

v0.10.22

March 09, 2021

Bug Fix

• Fix system metric logging rate in 0.10.x.
• Fix Audio external reference issue.
• Fix short runs with tensorboard_sync.
• Ignore wandb.init(id=) when running a sweep.
• Sanitize artifact metadata if needed.

Enhancement

• Allow syncing of tfevents with wandb sync --sync-tensorboard.

View the v0.10.22 changelog on GitHub.

​

v0.10.21

March 02, 2021

Bug Fix

• Fix artifact.get() regression since 0.10.18.
• Allow 0 byte artifacts.
• Fix codesaving and program name reporting.

Enhancement

• Added support for glb files for wandb.Object3D().
• Added support for external references for wandb.Audio().
• Custom chart support tensorboard pr_curves plugin.
• Support saving entire code directory in an artifact.

View the v0.10.21 changelog on GitHub.

​

v0.10.20

February 22, 2021

Bug Fix

• Wandb.login() now respects disabled mode.
• Handle exception when trying to log TPUs in colab.

Enhancement

• Add WANDB_START_METHOD=thread to support non-multiprocessing.
• Add group and job_type to Run object in the export API.
• Improve artifact docstrings.

View the v0.10.20 changelog on GitHub.

​

v0.10.19

February 14, 2021

Bug Fix

• Fix artifact manifest files incorrectly named with patch suffix.

View the v0.10.19 changelog on GitHub.

​

v0.10.18

February 08, 2021

Enhancement

• Add run delete and file delete to the public API.
• Align steps between tensorboard_sync and wandb.log() history.
• Add WANDB_START_METHOD to allow POSIX systems to use fork.
• Support mixed types in wandb.Table() with allow_mixed_types.

Bug Fix

• Fix potential leaked file due to log not being closed properly.
• Improve wandb verify to better handle network issues and report errors.
• Made file downloads more deterministic with respect to filesystem caches.

View the v0.10.18 changelog on GitHub.

​

v0.10.17

February 01, 2021

Bug Fix

• Fix regression seen with python 3.5.
• Silence vendored watchdog warnings on mac.

View the v0.10.17 changelog on GitHub.

​

v0.10.16

February 01, 2021

Enhancement

• Artifacts now support parallel writers for large distributed workflows.
• Artifacts support distributed tables for dataset visualization.
• Improvements to PR templates.
• Added more type annotations.
• Vendored watchdog 0.9.0 removing it as a dependency.
• New documentation generator.
• Public api now has file.direct_url to avoid redirects for signed urls.

Bug Fix

• Allow config-defaults.yaml to be overwritten when running sweeps.
• General bug fixes and improvements to wandb verify.
• Disabled widgets in Spyder IDE.
• Fixed WANDB_SILENT in Spyder IDE.
• Reference file:// artifacts respect the name attribute.

View the v0.10.16 changelog on GitHub.

​

v0.10.15

January 24, 2021

Enhancement

• Add wandb verify to troubleshoot local installs.

Bug Fix

• Fix tensorboard_sync issue writing to s3.
• Prevent git secrets from being stored.
• Disable verbose console messages when using moviepy.
• Fix artifacts with checkpoints to be more robust when overwriting files.
• Fix artifacts recycled id issue.

View the v0.10.15 changelog on GitHub.

​

v0.10.14

January 15, 2021

Enhancement

• Add wandb.Audio support to Artifacts.

Bug Fix

• Fix wandb config regressions introduced in 0.10.13.
• Rollback changes supporting media with slashes in keys.

View the v0.10.14 changelog on GitHub.

​

v0.10.13

January 11, 2021

Enhancement

• Add support for Mac M1 GPU monitoring.
• Add support for TPU monitoring.
• Add setting to disable sagemaker integration.

Bug Fix

• Fix tensorboard_sync with tensorboardX and tf1.
• Fix issues logging images with slashes.
• Fix custom charts issues.
• Improve error messages using wandb pull.
• Improve error messages with wandb.Table().
• Make sure silent mode is silent.
• Fix wandb online to renable logging.
• Multiple artifact fixes.

View the v0.10.13 changelog on GitHub.

​

v0.10.12

December 03, 2020

Enhancement

• Add Artifact.used_by and Artifact.logged_by.
• Validate type consistency when logging Artifacts.
• Enhance JoinedTable to not require downloaded assets.
• Add ability to recursively download dependent artifacts.
• Enable gradient logging with keras and tf2+.
• Validate pytorch models are passed to wandb.watch().
• Improved docstrings for public methods / objects.
• Warn when image sequences are logged with different sizes.

Bug Fix

• Fix incorrectly generated filenames in summary.
• Fix anonymous mode to include the api key in URLs.
• Fix pickle issue with disabled mode.
• Fix artifact from_id query.
• Fix handling of Tables with different image paths.

View the v0.10.12 changelog on GitHub.

​

v0.10.11

November 18, 2020

Enhancement

• Disable wandb logging with wandb disabled or wandb.init(mode="disabled").
• Support cloning an artifact when logging wandb.Image().

Bug Fix

• Multiple media artifact improvements and internal refactor.
• Improve handling of artifact errors.
• Fix issue where notebook name was ignored.
• Extend silent mode for jupyter logging.
• Fix issue where vendored libraries interfered with python path.
• Fix various exceptions (divide by zero, int conversion, TypeError).

View the v0.10.11 changelog on GitHub.

​

v0.10.10

November 09, 2020

Enhancement

• Added confusion matrix plot.
• Better jupyter messages with wandb.init()/reinit/finish.

Bug Fix

• Fix for fastai 2.1.5 (removed log_args).
• Fixed media logging when directories are changed.

View the v0.10.10 changelog on GitHub.

​

v0.10.9

November 04, 2020

Enhancement

• Added artifact media logging (alpha).
• Add scriptable alerts.
• Add url attribute for sweep public api.
• Update docstrings for wandb sdk functions.

Bug Fix

• Fix cases where offline mode was making network connections.
• Fix issues with python sweeps and run stopping.
• Fix logging issue where we could accidently display an api key.
• Fix wandb login issues with malformed hosts.
• Allow wandb.restore() to be called without wandb.init().
• Fix resuming (reusing run_id) with empty summary.
• Fix artitifact download issue.
• Add missing wandb.unwatch() function.
• Avoid creating spurious wandb directories.
• Fix collections import issue when using an old version of six.

View the v0.10.9 changelog on GitHub.

​

v0.10.8

October 22, 2020

Enhancement

• Allow callables to be serialized.

Bug Fix

• Fix compatibility issue with python 3.9.
• Fix wandb sync failure introduced in 0.10.6.
• Improve python agent handling of failing runs.
• Fix rare condition where resuming runs does not work.
• Improve symlink handling when called in thread context.
• Fix issues when changing directories before calling wandb.init().

View the v0.10.8 changelog on GitHub.

​

v0.10.7

October 15, 2020

Bug Fix

• Fix issue when checking for updated releases on pypi.

View the v0.10.7 changelog on GitHub.

​

v0.10.6

October 15, 2020

Bug Fix

• Make sure code saving is enabled in jupyter environments after login.
• Sweep agents have extended timeout for large sweep configs.
• Support WANDB_SILENT environment variable.
• Warn about missing python package when logging images.
• Fix wandb.restore() to apply diff patch.
• Improve artifact error messages.
• Fix loading of config-defaults.yaml and specified list of yaml config files.

View the v0.10.6 changelog on GitHub.

​

v0.10.5

October 07, 2020

Enhancement

• Add new custom plots: wandb.plot.*.
• Add new python based sweep agent: wandb.agent().

Bug Fix

• Console log fixes (tqdm on windows, fix close exceptions).
• Add more attributes to the Run object (group, job_type, urls).
• Fix sagemaker login issues.
• Fix issue where plots were not uploaded until the end of run.

View the v0.10.5 changelog on GitHub.

​

v0.10.4

September 29, 2020

Bug Fix

• Fix an issue where wandb.init(allow_val_change=) throws exception.

View the v0.10.4 changelog on GitHub.

​

v0.10.3

September 29, 2020

Enhancement

• Added warning when trying to sync pre 0.10.0 run dirs.
• Improved jupyter support for wandb run syncing information.

Bug Fix

• Fix artifact download issues.
• Fix multiple issues with tensorboard_sync.
• Fix multiple issues with juypter/python sweeps.
• Fix issue where login was timing out.
• Fix issue where config was overwritten when resuming runs.
• Ported sacred observer to 0.10.x release.
• Fix predicted bounding boxes overwritten by ground truth boxes.
• Add missing save_code parameter to wandb.init().

View the v0.10.3 changelog on GitHub.

​

v0.10.2

September 20, 2020

Enhancement

• Added upload_file to API.
• Wandb.finish() can be called without matching wandb.init().

Bug Fix

• Fix issue where files were being logged to wrong parallel runs.
• Fix missing properties/methods — as_dict(), sweep_id.
• Fix wandb.summary.update() not updating all keys.
• Code saving was not properly enabled based on UI settings.
• Tensorboard now logging images before end of program.
• Fix resume issues dealing with config and summary metrics.

View the v0.10.2 changelog on GitHub.

​

v0.10.1

September 16, 2020

Enhancement

• Added sync_tensorboard ability to handle S3 and GCS files.
• Added ability to specify host with login.
• Improved artifact API to allow modifying attributes.

Bug Fix

• Fix codesaving to respect the server settings.
• Fix issue running wandb.init() on restricted networks.
• Fix issue where we were ignoring settings changes.
• Fix artifact download issues.

View the v0.10.1 changelog on GitHub.

​

v0.10.0

September 11, 2020

Enhancement

• Added history sparklines at end of run.
• Artifact improvements and API for linking.
• Improved offline support and syncing.
• Basic noop mode support to simplify testing.
• Improved windows/pycharm support.
• Run object has more modifiable properties.
• Public API supports attaching artifacts to historic runs.

Bug Fix

• Many bugs fixed due to simplifying logic.

View the v0.10.0 changelog on GitHub.

​

v0.9.7

September 08, 2020

Enhancement

• New sacred observer available at wandb.sacred.WandbObserver.
• Improved artifact reference tracking for HTTP urls.

Bug Fix

• Print meaningful error message when runs are queried with summary instead of summary_metrics.

View the v0.9.7 changelog on GitHub.

​

v0.9.6

August 28, 2020

Enhancement

• Sub paths of artifacts now expose an optional root directory argument to download().
• Artifact.new_file accepts an optional mode argument.
• Removed legacy fastai docs as we’re now packaged with fastai v2!

Bug Fix

• Fix yaml parsing error handling logic.
• Bad spelling in torch docstring, thanks @mkkb473.

View the v0.9.6 changelog on GitHub.

​

v0.9.5

August 17, 2020

Enhancement

• Remove unused y_probas in sklearn plots, thanks @dreamflasher.
• New deletion apis for artifacts.

Bug Fix

• Fix wandb restore when not logged in.
• Fix artifact download paths on Windows.
• Retry 408 errors on upload.
• Fix mask numeric types, thanks @numpee.
• Fix artifact reference naming mixup.

View the v0.9.5 changelog on GitHub.

​

v0.9.4

July 24, 2020

Enhancement

• Default pytorch histogram logging frequency from 100 -> 1000 steps.

Bug Fix

• Fix multiple prompts for login when using the command line.
• Fix “no method rename_file” error.
• Fixed edgecase histogram calculation in PyTorch.
• Fix error in jupyter when saving session history.
• Correctly return artifact metadata in public api.
• Fix matplotlib / plotly rendering error.

View the v0.9.4 changelog on GitHub.

​

v0.9.3

July 10, 2020

Enhancement

• New artifact cli commands!

wandb artifact put path_file_or_ref
wandb artifact get artifact:version
wandb artifact ls project_name

• New artifact api commands!

wandb.log_artifact()
wandb.use_artifact()
wandb.Api().artifact_versions()
wandb.Api().run.used_artifacts()
wandb.Api().run.logged_artifacts()
wandb.Api().Artifact().file()

• Improved syncing of large wandb-history.jsonl files for wandb sync.
• New Artifact.verify method to ensure the integrity of local artifacts.
• Better testing harness for api commands.
• Run directory now store local time instead of utc time in the name, thanks @aiyolo!
• Improvements to our doc strings across the board.
• Wandb.Table now supports a dataframe argument for logging dataframes as tables!

Bug Fix

• Artifacts work in python2.
• Artifacts default download locations work in Windows.
• GCS references now properly cache / download, thanks @yoks!
• Fix encoding of numpy arrays to JSON.
• Fix string comparison error message.

View the v0.9.3 changelog on GitHub.

​

v0.9.2

June 29, 2020

Enhancement

• Major overhaul of artifact caching.
• Configurable cache directory for artifacts.
• Configurable download directory for artifacts.
• New Artifact.verify method to ensure the integrity of local artifacts.
• Use_artifact no longer requires type.
• Deleted artifacts can now be be recommitted.
• Lidar scenes now support vectors.

Bug Fix

• Fix issue with artifact downloads returning errors.
• Segmentation masks now handle non-unint8 data.
• Fixed path parsing logic in api.runs().

View the v0.9.2 changelog on GitHub.

​

v0.9.1

June 09, 2020

Bug Fix

• Fix issue where files were always logged to latest run in a project.
• Fix issue where url was not display url on first call to wandb.init.

View the v0.9.1 changelog on GitHub.

​

v0.9.0

June 05, 2020

Bug Fix

• Handle multiple inits in Jupyter.
• Handle ValueError’s when capturing signals, thanks @jsbroks.
• Wandb agent handles rate limiting properly.

Enhancement

• Wandb.Artifact is now generally available!
• Feature_importances now supports CatBoost, thanks @neomatrix369.

View the v0.9.0 changelog on GitHub.

​

v0.8.36

May 11, 2020

Bug Fix

• Catch all exceptions when saving Jupyter sessions.
• Validation_data automatically set in TF >= 2.2.
• implements* hooks now implemented in keras callback for TF >= 2.2.

Enhancement

• Raw source code saving now disabled by default.
• We now support global settings on boot to enable code saving on the server.
• New code_save=True argument to wandb.init to enable code saving manually.

View the v0.8.36 changelog on GitHub.

​

v0.8.35

May 01, 2020

Bug Fix

• Ensure cells don’t hang on completion.
• Fixed jupyter integration in PyCharm shells.
• Made session history saving handle None metadata in outputs.

View the v0.8.35 changelog on GitHub.

​

v0.8.34

April 28, 2020

Enhancement

• Save session history in jupyter notebooks.
• Kaggle internet enable notification.
• Extend wandb.plots.feature_importances to work with more model types, thanks @neomatrix369!

Bug Fix

• Code saving for jupyter notebooks restored.
• Fixed thread errors in jupyter.
• Ensure final history rows aren’t dropped in jupyter.

View the v0.8.34 changelog on GitHub.

​

v0.8.33

April 24, 2020

Enhancement

• Add default class labels for semantic segmentation.
• Enhance bounding box API to be similar to semantic segmentation API.

Bug Fix

• Increase media table rows to improve ROC/PR curve logging.
• Fix issue where pre binned histograms were not being handled properly.
• Handle nan values in pytorch histograms.
• Fix handling of binary image masks.

View the v0.8.33 changelog on GitHub.

​

v0.8.32

April 14, 2020

Enhancement

• Improve semantic segmentation image mask logging.

View the v0.8.32 changelog on GitHub.

​

v0.8.31

March 19, 2020

Enhancement

• Close all open files to avoice ResourceWarnings, thanks @CrafterKolyan!

Bug Fix

• Parse “tensor” protobufs, fixing issues with tensorboard syncing in 2.1.

View the v0.8.31 changelog on GitHub.

​

v0.8.30

March 19, 2020

Enhancement

• Add ROC, precision_recall, HeatMap, explainText, POS, and NER to wandb.plots.
• Add wandb.Molecule() logging.
• Capture kaggle runs for metrics.
• Add ability to watch from run object.

Bug Fix

• Avoid accidently picking up global debugging logs.

View the v0.8.30 changelog on GitHub.

​

v0.8.29

March 05, 2020

Enhancement

• Improve bounding box annotations.
• Log active GPU system metrics.
• Only writing wandb/settings file if wandb init is called.
• Improvements to wandb local command.

Bug Fix

• Fix GPU logging on some devices without power metrics.
• Fix sweep config command handling.
• Fix tensorflow string logging.

View the v0.8.29 changelog on GitHub.

​

v0.8.28

February 21, 2020

Enhancement

• Added code saving of main python module.
• Added ability to specify metadata for bounding boxes and segmentation masks.

Bug Fix

• Fix situations where uncommitted data from wandb.log() is not persisted.

View the v0.8.28 changelog on GitHub.

​

v0.8.27

February 11, 2020

Bug Fix

• Fix dependency conflict with new versions of six package.

View the v0.8.27 changelog on GitHub.

​

v0.8.26

February 10, 2020

Enhancement

• Add best metric and epoch to run summary with Keras callback.
• Added wandb.run.config_static for environments required pickled config.

Bug Fix

• Fixed regression causing failures with wandb.watch() and DataParallel.
• Improved compatibility with python 3.8.
• Fix model logging under windows.

View the v0.8.26 changelog on GitHub.

​

v0.8.25

February 04, 2020

Bug Fix

• Fix exception when using wandb.watch() in a notebook.
• Improve support for sparse tensor gradient logging on GPUs.

View the v0.8.25 changelog on GitHub.

​

v0.8.24

February 03, 2020

Bug Fix

• Relax version dependency for PyYAML for users with old environments.

View the v0.8.24 changelog on GitHub.

​

v0.8.23

February 03, 2020

Enhancement

• Added scikit-learn support.
• Added ability to specify/exclude specific keys when building wandb.config.

Bug Fix

• Fix wandb.watch() on sparse tensors.
• Fix incompatibilty with ray 0.8.1.
• Fix missing pyyaml requirement.
• Fix “W&B process failed to launch” problems.
• Improved ability to log large model graphs and plots.

View the v0.8.23 changelog on GitHub.

​

v0.8.22

January 24, 2020

Enhancement

• Added ability to configure agent commandline from sweep config.

Bug Fix

• Fix fast.ai prediction logging.
• Fix logging of eager tensorflow tensors.
• Fix jupyter issues with logging notebook name and wandb.watch().

View the v0.8.22 changelog on GitHub.

​

v0.8.21

January 15, 2020

Enhancement

• Ignore wandb.init() specified project and entity when running a sweep.

Bug Fix

• Fix agent “flapping” detection.
• Fix local controller not starting when sweep is pending.

View the v0.8.21 changelog on GitHub.

​

v0.8.20

January 10, 2020

Enhancement

• Added support for LightGBM.
• Added local board support (Experimental).
• Added ability to modify sweep configuration.
• Added GPU power logging to system metrics.

Bug Fix

• Prevent sweep agent from failing continuously when misconfigured.

View the v0.8.20 changelog on GitHub.

​

v0.8.19

December 18, 2019

Enhancement

• Added beta support for ray/tune hyperopt search strategy.
• Added ability to specify max runs per agent.
• Improve experience starting a sweep without a project already created.

Bug Fix

• Fix repeated wandb.Api().Run(id).scan_history() calls get updated data.
• Fix early_terminate/hyperband in notebook/python environments.

View the v0.8.19 changelog on GitHub.

​

v0.8.18

December 04, 2019

Enhancement

• Added min_step and max_step to run.scan_history for grabbing sub-sections of metrics.
• Wandb.init(reinit=True) now automatically calls wandb.join() to better support multiple runs per process.

Bug Fix

• Wandb.init(sync_tensorboard=True) works again for TensorFlow 2.0.

View the v0.8.18 changelog on GitHub.

​

v0.8.17

December 02, 2019

Enhancement

• Handle tags being passed in as a string.

Bug Fix

• Pin graphql-core < 3.0.0 to fix install errors.
• TQDM progress bars update logs properly.
• Oversized summary or history logs are now dropped which prevents retry hanging.

View the v0.8.17 changelog on GitHub.

​

v0.8.16

November 21, 2019

Bug Fix

• Fix regression syncing some versions of Tensorboard since 0.8.13.
• Fix network error in Jupyter.

View the v0.8.16 changelog on GitHub.

​

v0.8.15

November 05, 2019

Bug Fix

• Fix calling wandb.init with sync_tensorboard multiple times in Jupyter.
• Fix RuntimeError race when using threads and calling wandb.log.
• Don’t initialize Sentry when error reporting is disabled.

Enhancement

• Added best_run() to wandb.sweep() public Api objects.
• Remove internal tracking keys from wandb.config objects in the public Api.

View the v0.8.15 changelog on GitHub.

​

v0.8.14

November 01, 2019

Bug Fix

• Improve large object warning when values reach maximum size.
• Warn when wandb.save isn’t passed a string.
• Run stopping from the UI works since regressing in 0.8.12.
• Restoring a file that already exists locally works.
• Fixed TensorBoard incorrectly placing some keys in the wrong step since 0.8.10.
• Wandb.Video only accepts uint8 instead of incorrectly converting to floats.
• SageMaker environment detection is now more robust.
• Resuming correctly populates config.
• Wandb.restore respects root when run.dir is set #658.
• Calling wandb.watch multiple times properly namespaces histograms and graphs.

Enhancement

• Sweeps now work in Windows!
• Added sweep attribute to Run in the public api.
• Added sweep link to Jupyter and terminal output.
• TensorBoard logging now stores proper timestamps when importing historic results.
• TensorBoard logging now supports configuring rate_limits and filtering event types.
• Use simple output mirroring stdout doesn’t have a file descriptor.
• Write wandb meta files to the system temp directory if the local directory isn’t writable.
• Added beta api.reports to the public API.
• Added wandb.unwatch to remove hooks from pytorch models.
• Store the framework used in config._wandb.

View the v0.8.14 changelog on GitHub.

​

v0.8.13

October 15, 2019

Bug Fix

• Create nested directory when videos are logged from tensorboard namespaces.
• Fix race when using wandb.log async=True.
• Run.summary acts like a proper dictionary.
• Run.summary sub dictionaries properly render.
• Handle None when passing class_colors for segmentation masks.
• Handle tensorflow2 not having a SessionHook.
• Properly escape args in windows.
• Fix hanging login when in anonymode.
• Tf2 keras patch now handles missing callbacks args.

Enhancement

• Updates documentation autogenerated from docstrings in /docs.
• Wandb.init(config=config_dict) does not update sweep specified parameters.
• Wandb.config object now has a setdefaults method enabling improved sweep support.
• Improved terminal and jupyter message incorporating emojii!
• Allow wandb.watch to be called multiple times on different models.
• Improved support for watching multiple tfevent files.
• Windows no longer requires wandb run simply run python script_name.py.
• wandb agent now works on windows.
• Nice error message when wandb.log is called without a dict.
• Keras callback has a new log_batch_frequency for logging metrics every N batches.

View the v0.8.13 changelog on GitHub.

​

v0.8.12

September 20, 2019

Bug Fix

• Fix compatibility issue with python 2.7 and old pip dependencies.

Enhancement

• Improved onboarding flow when creating new accounts and entering api_key.

View the v0.8.12 changelog on GitHub.

​

v0.8.11

September 19, 2019

Bug Fix

• Fix public api returning incorrect data when config value is 0 or False.
• Resumed runs no longer overwrite run names with run id.

Enhancement

• Added recording of spell.run id in config.

View the v0.8.11 changelog on GitHub.

​

v0.8.10

September 13, 2019

Bug Fix

• Wandb magic handles the case of tf.keras and keras being loaded.
• Tensorboard logging won’t drop steps if multiple loggers have different global_steps.
• Keras gradient logging works in the latest tf.keras.
• Keras validation_data is properly set in tensorflow 2.
• Wandb pull command creates directories if they don’t exist, thanks @chmod644.
• File upload batching now asserts a minimum size.
• Sweeps works in python2 again.
• Scan_history now iterates the full set of points.
• Jupyter will run local mode if credentials can’t be obtained.

Enhancement

• Sweeps can now be run from within jupyter / directly from python! https://docs.wandb.com/sweeps/python.
• New openai gym integration will automatically log videos, enabled with the monitor_gym keyword argument to wandb.init.
• Ray Tune logging callback in wandb.ray.WandbLogger.
• New global config file in ~/.config/wandb for global settings.
• Added tests for fastai, thanks @borisdayma.
• Public api performance enhancements.
• Deprecated username in favor of entity in the public api for consistency.
• Anonymous login support enabled by default.
• New wandb.login method to be used in jupyter enabling anonymous logins.
• Better dependency error messages for data frames.
• Initial integration with spell.run.
• All images are now rendered as PNG to avoid JPEG artifacts.
• Public api now has a projects field.

View the v0.8.10 changelog on GitHub.

​

v0.8.9

August 19, 2019

Bug Fix

• Run.summary updates work in jupyter before log is called.
• Don’t require numpy to be installed.
• Setting nested keys in summary works.
• Notebooks in nested directories are properly saved.
• Don’t retry 404’s / better error messaging from the server.
• Strip leading slashes when loading paths in the public api.

Enhancement

• Small files are batch uploaded as gzipped tarballs.
• TensorBoardX gifs are logged to wandb.

View the v0.8.9 changelog on GitHub.

​

v0.8.8

August 13, 2019

Bug Fix

• Wandb.init properly handles network failures on startup.
• Keras callback only logs examples if data_type or input_type is set.
• Fix edge case PyTorch model logging bug.
• Handle patching tensorboard multiple times in jupyter.
• Sweep picks up config.yaml from the run directory.
• Dataframes handle integer labels.
• Handle invalid JSON when querying jupyter servers.

Enhancement

• Fastai uses a fixed seed for example logging.
• Increased the max number of images for fastai callback.
• New wandb.Video tag for logging video.
• Sync=False argument to wandb.log moves logging to a thread.
• New local sweep controller for custom search logic.
• Anonymous login support for easier onboarding.
• Calling wandb.init multiple times in jupyter doesn’t error out.

View the v0.8.8 changelog on GitHub.

​

v0.8.7

August 07, 2019

Bug Fix

• Keras callback no longer guesses input_type for 2D data.
• Wandb.Image handles images with 1px height.

Enhancement

• Wandb Public API now has run.scan_history to return all history rows.
• Wandb.config prints helpful errors if used before calling init.
• Wandb.summary prints helpful errors if used before calling init.
• Filestream api points to new url on the backend.

View the v0.8.7 changelog on GitHub.

​

v0.8.6

July 31, 2019

Bug Fix

• Fastai callback uses the default monitor instead of assuming val_loss.
• Notebook introspections handles error cases and doesn’t print stacktrace on failure.
• Don’t print description warning when setting name.
• Fixed dataframe logging error with the keras callback.
• Fixed line offsets in logs when resuming runs.
• Wandb.config casts non-builtins before writing to yaml.
• Vendored backports.tempfile to address missing package on install.

Enhancement

• Added api.sweep to the python export api for querying sweeps.
• Added WANDB_NOTEBOOK_NAME for specifying the notebook name in cases we can’t infer it.
• Added WANDB_HOST to override hostnames.
• Store if a run was run within jupyter.
• Wandb now supports stopping runs from the web ui.
• Handle floats passed as step to wandb.log.
• Wandb.config has full unicode support.
• Sync the main file to wandb if code saving is enabled and it’s untracked by git.
• XGBoost callback: wandb.xgboost.wandb_callback().

View the v0.8.6 changelog on GitHub.

​

v0.8.5

July 12, 2019

Bug Fix

• Fixed plotly charts with large numpy arrays not rendering.
• wandb docker works when nvidia is present.
• Better error when non string keys are sent to log.
• Relaxed pyyaml dependency to fix AMI installs.
• Magic works in jupyter notebooks.

Enhancement

• New preview release of auto-dataframes for Keras.
• Added input_type and output_type to the Keras callback for simpler config.
• Public api supports retrieving specific keys and custom xaxis.

View the v0.8.5 changelog on GitHub.

​

v0.8.4

July 08, 2019

Bug Fix

• WANDB_IGNORE_GLOBS is respected on the final scan of files.
• Unified run.id, run.name, and run.notes across all apis.
• Handle funky terminal sizes when setting up our pseudo tty.
• Fixed Jupyter notebook introspection logic.
• Run.summary.update() persists changes to the server.
• Tensorboard syncing is robust to invalid histograms and truncated files.

Enhancement

• Preview release of magic, calling wandb.init(magic=True) should automatically track config and metrics when possible.
• Cli now supports local installs of the backend.
• Fastai callback supports logging example images.

View the v0.8.4 changelog on GitHub.

​

v0.8.3

June 26, 2019

Bug Fix

• Image logging works in Windows.
• Wandb sync handles tfevents with a single timestep.
• Fix incorrect command in overview page for running runs.
• Handle histograms with > 512 bins when streaming tensorboard.
• Better error message when calling wandb sync on a file instead of a directory.

Enhancement

• New helper function for handling hyperparameters in sweeps wandb.config.user_items().
• Better mocking for improved testing.

View the v0.8.3 changelog on GitHub.

​

v0.8.2

June 20, 2019

Bug Fix

• Entity is persisted on wandb.run when queried from the server.
• Tmp files always use the temporary directory to avoid syncing.
• Raise error if file shrinks while uploading.
• Images log properly in windows.
• Upgraded pyyaml requirement to address CVE.
• No longer store a history of rows to prevent memory leak.

Enhancement

• Summary now supports new dataframe format.
• WANDB_SILENT environment variable writes all wandb messages to debug.log.
• Improved error messages for windows and tensorboard logging.
• Output.log is uploaded at the end of each run.
• Metadata, requirements, and patches are uploaded at the beginning of a run.
• When not running from a git repository, store the main python file.
• Added WANDB_DISABLE_CODE to prevent diffing and code saving.
• When running in jupyter store the name of the notebook.
• Auto-login support for colab.
• Store url to colab notebook.
• Store the version of this library in config.
• Store sys.executable in metadata.
• Fastai callback no longer requires path.
• Wandb.init now accepts a notes argument.
• The cli replaced the message argument with notes and name.

View the v0.8.2 changelog on GitHub.

​

v0.8.1

May 23, 2019

Bug Fix

• Wandb sync handles tensorboard embeddings.
• Wandb sync correctly handles images in tensorboard.
• Tf.keras correctly handles single input functional models.
• Wandb.Api().runs returns an iterator that’s reusable.
• WANDB_DIR within a hidden directory doesn’t prevent syncing.
• Run.files() iterates over all files.
• Pytorch recursion too deep error.

Enhancement

• Wandb sync accepts an —ignore argument with globs to skip files.
• Run.summary now has an items() method for iterating over all keys.

View the v0.8.1 changelog on GitHub.

​

v0.8.0

May 17, 2019

Bug Fix

• Better error messages on access denied.
• Better error messages when optional packages aren’t installed.
• Urls printed to the terminal are url-escaped.
• Namespaced tensorboard events work with histograms.
• Public API now retries on failures and re-uses connection pool.
• Catch git errors when remotes aren’t pushed to origin.
• Moved keras graph collection to on_train_begin to handle unbuilt models.
• Handle more cases of not being able to save weights.
• Updates to summary after resuming are persisted.
• PyTorch histc logging fixed in 0.4.1.
• Fixed wandb sync tensorboard import.

Enhancement

• Wandb.init(tensorboard=True) works with Tensorflow 2 and Eager Execution.
• Wandb.init(tensorboard=True) now works with tb-nightly and PyTorch.
• Automatically log examples with tf.keras by adding missing validation_data.
• Socket only binds to localhost for improved security and prevents firewall warnings in OSX.
• Added user object to public api for getting the source user.
• Added run.display_name to the public api.
• Show display name in console output.
• Added —tags, —job_group, and —job_type to wandb run.
• Added environment variable for minimum time to run before considering crashed.
• Added flake8 tests to CI, thanks @cclauss!

View the v0.8.0 changelog on GitHub.

​

v0.7.3

April 15, 2019

Bug Fix

• Wandb-docker-run accepts image digests.
• Keras callback works in tensorflow2-alpha0.
• Keras model graph now puts input layer first.

Enhancement

• PyTorch log frequency added for gradients and weights.
• PyTorch logging performance enhancements.
• Wandb.init now accepts a name parameter for naming runs.
• Wandb.run.name reflects custom display names.
• Improvements to nested summary values.
• Deprecated wandb.Table.add_row in favor of wandb.Table.add_data.
• Initial support for a fast.ai callback thanks to @borisdayma!

View the v0.7.3 changelog on GitHub.

​

v0.7.2

March 19, 2019

Bug Fix

• Run.get_url resolves the default entity if one wasn’t specified.
• Wandb restore accepts run paths with only slashes.
• Fixed PyYaml deprecation warnings.
• Added entrypoint shell script to manifest.
• Strip newlines from cuda version.

View the v0.7.2 changelog on GitHub.

​

v0.7.1

March 14, 2019

Bug Fix

• Handle case insensitive docker credentials.
• Fix app_url for private cloud login flow.
• Don’t retry 404’s when starting sweep agents.

View the v0.7.1 changelog on GitHub.

​

v0.7.0

February 28, 2019

Bug Fix

• Ensure DNS lookup failures can’t prevent startup.
• Centralized debug logging.
• Wandb agent waits longer to send a SIGKILL after sending SIGINT.

Enhancement

• Support for logging docker images with the WANDB_DOCKER env var.
• WANDB_DOCKER automatically set when run in kubernetes.
• New wandb-docker-run command to automatically set env vars and mount code.
• Wandb.restore supports launching docker for runs that ran with it.
• Python packages are now recorded and saved in a requirements.txt file.
• Cpu_count, gpu_count, gpu, os, and python version stored in wandb-metadata.json.
• The export api now supports docker-like paths, i.e. username/project:run_id.
• Better first time user messages and login info.

View the v0.7.0 changelog on GitHub.

​

v0.6.35

January 29, 2019

Bug Fix

• Improve error reporting for sweeps.

View the v0.6.35 changelog on GitHub.

​

v0.6.35

January 29, 2019

Bug Fix

• Improve error reporting for sweeps.

View the v0.6.35 changelog on GitHub.

​

v0.6.35

January 29, 2019

Bug Fix

• Improve error reporting for sweeps.

View the v0.6.35 changelog on GitHub.

​

v0.6.35

January 29, 2019

Bug Fix

• Improve error reporting for sweeps.

View the v0.6.35 changelog on GitHub.

​

v0.6.35

January 29, 2019

Bug Fix

• Improve error reporting for sweeps.

View the v0.6.35 changelog on GitHub.

​

v0.6.35

January 29, 2019

Bug Fix

• Improve error reporting for sweeps.

View the v0.6.35 changelog on GitHub.

​

v0.6.34

January 23, 2019

Bug Fix

• Fixed Jupyter logging, don’t change logger level.
• Fixed resuming in Jupyter.

Enhancement

• Wandb.init now degrades gracefully if a user hasn’t logged in to wandb.
• Added a force flag to wandb.init to require a machine to be logged in.
• Tensorboard and TensorboardX logging is now automatically instrumented when enabled.
• Added a tensorboard to wandb.init which patches tensorboard for logging.
• Wandb.save handles now accepts a base path to files in sub directories.
• Wandb.tensorflow and wandb.tensorboard can now be accessed without directly importing.
• wandb sync will now traverse a wandb run directory and sync all runs.

View the v0.6.34 changelog on GitHub.

​

v0.6.34

January 23, 2019

Bug Fix

• Fixed Jupyter logging, don’t change logger level.
• Fixed resuming in Jupyter.

Enhancement

• Wandb.init now degrades gracefully if a user hasn’t logged in to wandb.
• Added a force flag to wandb.init to require a machine to be logged in.
• Tensorboard and TensorboardX logging is now automatically instrumented when enabled.
• Added a tensorboard to wandb.init which patches tensorboard for logging.
• Wandb.save handles now accepts a base path to files in sub directories.
• Wandb.tensorflow and wandb.tensorboard can now be accessed without directly importing.
• wandb sync will now traverse a wandb run directory and sync all runs.

View the v0.6.34 changelog on GitHub.

​

v0.6.34

January 23, 2019

Bug Fix

• Fixed Jupyter logging, don’t change logger level.
• Fixed resuming in Jupyter.

Enhancement

• Wandb.init now degrades gracefully if a user hasn’t logged in to wandb.
• Added a force flag to wandb.init to require a machine to be logged in.
• Tensorboard and TensorboardX logging is now automatically instrumented when enabled.
• Added a tensorboard to wandb.init which patches tensorboard for logging.
• Wandb.save handles now accepts a base path to files in sub directories.
• Wandb.tensorflow and wandb.tensorboard can now be accessed without directly importing.
• wandb sync will now traverse a wandb run directory and sync all runs.

View the v0.6.34 changelog on GitHub.

​

v0.6.34

January 23, 2019

Bug Fix

• fixed Jupyter logging, don’t change logger level.
• fixed resuming in Jupyter.

Enhancement

• wandb.init now degrades gracefully if a user hasn’t logged in to wandb.
• added a force flag to wandb.init to require a machine to be logged in.
• Tensorboard and TensorboardX logging is now automatically instrumented when enabled.
• added a tensorboard to wandb.init which patches tensorboard for logging.
• wandb.save handles now accepts a base path to files in sub directories.
• wandb.tensorflow and wandb.tensorboard can now be accessed without directly importing.
• wandb sync will now traverse a wandb run directory and sync all runs.

View the v0.6.34 changelog on GitHub.

​

v0.6.34

January 23, 2019

Bug Fix

• Fixed Jupyter logging, don’t change logger level.
• Fixed resuming in Jupyter.

Enhancement

• Wandb.init now degrades gracefully if a user hasn’t logged in to wandb.
• Added a force flag to wandb.init to require a machine to be logged in.
• Tensorboard and TensorboardX logging is now automatically instrumented when enabled.
• Added a tensorboard to wandb.init which patches tensorboard for logging.
• Wandb.save handles now accepts a base path to files in sub directories.
• Wandb.tensorflow and wandb.tensorboard can now be accessed without directly importing.
• wandb sync will now traverse a wandb run directory and sync all runs.

View the v0.6.34 changelog on GitHub.

​

v0.6.34

January 23, 2019

Bug Fix

• Fixed Jupyter logging, don’t change logger level.
• Fixed resuming in Jupyter.

Enhancement

• Wandb.init now degrades gracefully if a user hasn’t logged in to wandb.
• Added a force flag to wandb.init to require a machine to be logged in.
• Tensorboard and TensorboardX logging is now automatically instrumented when enabled.
• Added a tensorboard to wandb.init which patches tensorboard for logging.
• Wandb.save handles now accepts a base path to files in sub directories.
• Wandb.tensorflow and wandb.tensorboard can now be accessed without directly importing.
• wandb sync will now traverse a wandb run directory and sync all runs.

View the v0.6.34 changelog on GitHub.

​

v0.6.33

January 22, 2019

Bug Fix

• Fixed race where wandb process could hang at the end of a run.

View the v0.6.33 changelog on GitHub.

​

v0.6.33

January 22, 2019

Bug Fix

• Fixed race where wandb process could hang at the end of a run.

View the v0.6.33 changelog on GitHub.

​

v0.6.33

January 22, 2019

Bug Fix

• Fixed race where wandb process could hang at the end of a run.

View the v0.6.33 changelog on GitHub.

​

v0.6.33

January 22, 2019

Bug Fix

• Fixed race where wandb process could hang at the end of a run.

View the v0.6.33 changelog on GitHub.

​

v0.6.33

January 22, 2019

Bug Fix

• Fixed race where wandb process could hang at the end of a run.

View the v0.6.33 changelog on GitHub.

​

v0.6.33

January 22, 2019

Bug Fix

• Fixed race where wandb process could hang at the end of a run.

View the v0.6.33 changelog on GitHub.

​

v0.6.32

December 22, 2018

Bug Fix

• Fix resuming in Jupyter on kernel restart.
• Wandb.save ensures files are pushed regardless of growth.

Enhancement

• Added replace=True keyword to init for auto-resuming.
• New run.resumed property that can be used to detect if we’re resuming.
• New run.step property to use for setting an initial epoch on resuming.
• Made Keras callback save the best model as it improves.

View the v0.6.32 changelog on GitHub.

​

v0.6.31

December 20, 2018

Bug Fix

• Really don’t require numpy.
• Better error message if wandb.log is called before wandb.init.
• Prevent calling wandb.watch multiple times.
• Handle datetime attributes in logs / plotly.

Enhancement

• Add environment to sweeps.
• Enable tagging in the public API and in wandb.init.
• New media type wandb.Html for logging arbitrary html.
• Add Public api.create_run method for custom integrations.
• Added glob support to wandb.save, files save as they’re written to.
• Added wandb.restore for pulling files on resume.

View the v0.6.31 changelog on GitHub.

​

v0.6.30

December 06, 2018

Bug Fix

• Added a timeout for generating diffs on large repos.
• Fixed edge case where file syncing could hang.
• Ensure all file changes are captured before exit.
• Handle cases of sys.exit where code isn’t passed.
• Don’t require numpy.

Enhancement

• New wandb sync command that pushes a local directory to the cloud.
• Support for syncing tfevents file during training.
• Detect when running as TFJob and auto group.
• New Kubeflow module with initial helpers for pipelines.

View the v0.6.30 changelog on GitHub.

​

v0.6.29

November 26, 2018

Bug Fix

• Fixed history / summary bug.

View the v0.6.29 changelog on GitHub.

​

v0.6.28

November 24, 2018

Enhancement

• Initial support for AWS SageMaker.
• hook_torch renamed to watch with a deprecation warning.
• Projects are automatically created if they don’t exist.
• Additional GPU memory_allocated metric added.
• Keras Graph stores edges.

Bug Fix

• PyTorch graph parsing is more robust.
• Fixed PyTorch 0.3 support.
• File download API supports WANDB_API_KEY authentication.

View the v0.6.28 changelog on GitHub.

​

v0.6.27

November 13, 2018

Enhancement

• Sweeps work with new backend (early release).
• Summary tracks all history metrics unless they’re overridden by directly writing. to summary.
• Files support in data API.

Bug Fix

• Show ongoing media file uploads in final upload progress.

View the v0.6.27 changelog on GitHub.

​

v0.6.26

November 09, 2018

Enhancement

• Wandb.Audio supports duration.

Bug Fix

• Pass username header in filestream API.

View the v0.6.26 changelog on GitHub.

​

v0.6.25

November 08, 2018

Enhancement

• New wandb.Audio data type.
• New step keyword argument when logging metrics.
• Ability to specify run group and job type when calling wandb.init() or via. environment variables. This enables automatic grouping of distributed training runs in the UI
• Ability to override username when using a service account API key.

Bug Fix

• Handle non-tty environments in Python2.
• Handle non-existing git binary.
• Fix issue where sometimes the same image was logged twice during a Keras step.

View the v0.6.25 changelog on GitHub.

​

v0.6.23

October 19, 2018

Enhancement

• PyTorch.
  • Added a new wandb.hook_torch method which records the graph and logs gradients & parameters of pytorch models
  • wandb.Image detects pytorch tensors and uses torchvision.utils.make_grid to render the image.

Bug Fix

• wandb restore handles the case of not being run from within a git repo.

View the v0.6.23 changelog on GitHub.

​

v0.6.22

October 18, 2018

Bug Fix

• We now open stdout and stderr in raw mode in Python 2 ensuring tools like bpdb work.

View the v0.6.22 changelog on GitHub.

​

v0.6.21

October 12, 2018

Enhancement

• Catastrophic errors are now reported to Sentry unless WANDB_ERROR_REPORTING is set to false.
• Improved error handling and messaging on startup.

View the v0.6.21 changelog on GitHub.

​

v0.6.20

October 05, 2018

Bug Fix

• The first image when calling wandb.log was not being written, now it is.
• wandb.log and run.summary now remove whitespace from keys.

View the v0.6.20 changelog on GitHub.

​

v0.6.19

October 05, 2018

Bug Fix

• Vendored prompt_toolkit < 1.0.15 because the latest ipython is pinned > 2.0.
• Lazy load wandb.h5 only if summary is accessed to improve Data API performance.

Enhancement

• Jupyter.
  • Deprecated wandb.monitor in favor of automatically starting system metrics after the first wandb.log call
  • Added new %%wandb jupyter magic method to display live results
  • Removed jupyter description iframe
• The Data API now supports per_page and order options to the api.runs method.
• Initial support for wandb.Table logging.
• Initial support for matplotlib logging.

View the v0.6.19 changelog on GitHub.