v0.11.1

• Support different runs having different metadata in get_runs dataframe output.
• (Bug fix) Error message when creating a duplicate run is now more clear.
• (Bug fix) Correction to stopping the worker thread in situations where the run never started.

v0.11.0

• Support optional dataframe output from get_runs.

v0.10.1

• The worker process now no longer gives a long delay when a run has finished (now at most ~1 second).
• The worker process ends when the Run() context ends or close is called, rather than only when the main process exits.

v0.10.0

• The client class can now be used to retrieve runs.

v0.9.1

• (Bug fix) Retries in POST/PUTs to REST APIs didn't happen.
• Warn users if allow_pickle=True is required.

v0.9.0

• Set status to failed or terminated if the context manager is used and there is an exception.

v0.8.0

• Support NumPy arrays, PyTorch tensors, Matplotlib and Plotly plots and picklable Python objects as artifacts.
• (Bug fix) Events in offline mode didn't work.

v0.7.2

• Pydantic model is used for input validation.
• Support NaN, -inf and inf in metadata and metrics.

v0.7.0

• Collect CPU, GPU and memory resource metrics.
• Automatically delete temporary files used in offline mode once runs have entered a terminal state.
• Warn users if their access token has expired.
• Remove dependency on the randomname module, instead handle name generation server side.

v0.6.0

• offline and disabled options replaced with single mode flag