Skip to main content

Version 2.1

ClearML 2.1.9​

Bug fixes and Improvements

  • Add method to create a Hyper-Dataset version snapshot
  • Implement static route validator RouterService class
  • Fix syntax-related error in the Dataset.get() (#1648)
  • Improvements in formatting via f-strings instead of .format

ClearML 2.1.8​

New Features Improvements

  • HyperDatasets
    • Add ability to publish a HyperDataset via single method call
    • Add HyperDataset tagging
  • Datasets
    • Add path traversal check in Dataset._merge_datasets
    • Add status attribute in Dataset.list_datasets dataset return values
  • Add support for DefaultAzureCredential

Bug Fix

  • Fix issue that users restricted to a project cannot create Dataviews

ClearML 2.1.7​

Bug fixes and Improvements

  • Add optional (opt-out) parameter to prevent processing of pickled artifacts (#1619, #1621):
    • Use task.artifacts[0].get(block_unsafe_artifacts=True), or
    • Via configuration with the environment variable CLEARML_BLOCK_PICKLED_ARTIFACTS=1
    • Via sdk.storage.block_pickled_artifacts: true
  • Prevent resource leak during interrupted queries in GPUStatCollection._new_query_nvidia (#1617)
  • Add vulnerability check in Task.import_offline_session when extracting .zip archives to prevent path-traversal attacks (#1620)

ClearML 2.1.6​

Bug Fixes and Improvements

  • Add default shell binary option for script execution (#1586)
  • Add option to set plots upload destination (#1587)
  • Add exc_info param to improve error logging (#1550)
  • Raise duplicate InputModel import log from debug to info (#1582)
  • Add pickle file integrity hash verification for pd.DataFrame pickled artifacts (#1599)
  • Add check to avoid KeyError in dataset (#1598)
  • Routing fixes when using queues on autoscaler instances (#1602)
  • Refactor clearml/storage
  • Add hash comparison for external link change detection, subset file pulling in Dataset.get_local_copy(), and stale link entry pruning in Dataset.sync_folder (#1611)

ClearML 2.1.5​

Bug fixes and improvements

  • Add task method to remove tags
  • Fix _SaveFramesRequestNoValidate in offline mode (#1517)

ClearML 2.1.4​

New Features and Bug Fixes

  • Fix syntax error on Python 3.14 (#1521)
  • Add support for boto3 S3 specific configuration (#1516)
  • Add option to close squashed dataset (#1513)
  • Move None check earlier in CacheContext.get_local_copy (#1570)
  • Update to examples/hyperdatasets/finetune_qa_lora.py
  • Fix race condition with not using incremental logging config
  • Validate queue visibility in job scheduler
  • Use UTC everywhere in clearml/automation/scheduler.py
  • Implement improvement in is_within_directory

ClearML 2.1.3​

New Features and Bug Fixes

  • Fix GPU reporting for NVIDIA_VISIBLE_DEVICES=void
  • Fix default example parameters for sklearn joblib
  • Add support for ClearML App Gateway static routes in Gradio binding

ClearML 2.1.2​

Bug Fix

  • Fix broken ArgParser integration with SUPPRESS

ClearML 2.1.1​

New Features and Bug Fixes

  • Fix space is missing from the safe characters list when quoting downloaded file names
  • Add support for sdk.storage.http.legacy_fileservers to allow downloading data from legacy fileservers
  • Add Python 3.14 support

ClearML 2.1.0​

New Features

  • Add Hyper-Datasets support (Enterprise server required)
  • Update datetime usage (#1491)
  • Remove support for Python 3.5 and lower in required packages
  • Add support for multiple ports/endpoints in Task.request_external_endpoint() and router (Enterprise server required)
  • Improve GPU reporting on ARM GPUs
  • Add model fine-tuning and model embedding examples (#1483)

Bug Fixes

  • Fix Task.init() fails for tags with tuple type (#1468)
  • Make sure git diffs are always valid inputs for git apply (#1479)
  • Fix duplicate files in dataset uploads (#1463)
  • Fix transient error in machine stats should not break reporting