telemetry.callbacks

telemetry.callbacks

Trainer callbacks for reporting runtime metrics at regular intervals.

Classes

Name	Description
TelemetryCallback	Trainer callback for tracking and reporting runtime metrics.

telemetry.callbacks.TelemetryCallback()

Trainer callback for tracking and reporting runtime metrics.

This callback tracks training progress, runtime, and memory usage, sending telemetry at configurable intervals.

Name	Description
on_epoch_begin	Handle epoch start.
on_epoch_end	Handle epoch end.
on_log	Record training metrics with the step at which they were logged.
on_prediction_step	Capture evaluation peaks before the memory logger resets them.
on_step_end	Handle step end.
on_train_begin	Handle training start.
on_train_end	Handle training end.

telemetry.callbacks.TelemetryCallback.on_epoch_begin(
    args,
    state,
    control,
    **kwargs,
)

Handle epoch start.

telemetry.callbacks.TelemetryCallback.on_epoch_end(
    args,
    state,
    control,
    **kwargs,
)

Handle epoch end.

telemetry.callbacks.TelemetryCallback.on_log(
    args,
    state,
    control,
    logs=None,
    **kwargs,
)

Record training metrics with the step at which they were logged.

telemetry.callbacks.TelemetryCallback.on_prediction_step(
    args,
    state,
    control,
    **kwargs,
)

Capture evaluation peaks before the memory logger resets them.

telemetry.callbacks.TelemetryCallback.on_step_end(
    args,
    state,
    control,
    **kwargs,
)

Handle step end.

telemetry.callbacks.TelemetryCallback.on_train_begin(
    args,
    state,
    control,
    **kwargs,
)

Handle training start.

telemetry.callbacks.TelemetryCallback.on_train_end(
    args,
    state,
    control,
    **kwargs,
)

Handle training end.