kempnerforge.metrics.mlflow¶
MLflow experiment tracking: the training backend, checkpoint models and the eval publisher.
backend holds the logging backend MetricsTracker builds on rank 0 when
metrics.enable_mlflow is on, and worker the threads that bound each of its server calls in
time. auth prepares the environment Databricks auth reads, models finds or makes the logged
model of a checkpoint, and publish appends eval metrics to the run a checkpoint was trained
under (python -m kempnerforge.metrics.mlflow.publish). mlflow is imported only by the calls
that need it, so none of these modules requires it at import time.
Modules
Databricks auth for MLflow: the environment the SDK and artifact clients read. |
|
The MLflow logging backend, for Databricks or any MLflow tracking server. |
|
The MLflow logged model of a checkpoint: one external model per run and step. |
|
Append eval metrics to the MLflow run a checkpoint was trained under. |
|
Tracking calls bounded in time, so a slow or unreachable server never stalls training. |