SwanLab

merge_callbacks ​

python
def merge_callbacks(
    callbacks: CallbacksType,
) -> None:
ParameterTypeDescription
callbacksCallbacksTypeA single Callback object or an iterable of Callback objects to merge into the global registry.

Introduction ​

swanlab.merge_callbacks() merges custom callbacks into the global SwanLab callback registry. This allows you to register callbacks before calling swanlab.init(), enabling plugin injection in frameworks where swanlab.init() is not directly accessible (e.g., when using SwanLab with Transformers).

TIP

merge_callbacks must be called before swanlab.init(). It will raise a RuntimeError if called while a run is active.

Usage ​

Single Callback ​

python
import swanlab
from swanlab import Callback

class MyCallback(Callback):
    @property
    def name(self):
        return "my_callback"

    def on_run_initialized(self, run_dir, path, **kwargs):
        print("Run initialized!")

swanlab.merge_callbacks(MyCallback())
swanlab.init(project="my-project")

Multiple Callbacks ​

Pass a list to merge multiple callbacks at once:

python
import swanlab
from swanlab import Callback

class MyCallback(Callback):
    @property
    def name(self):
        return "my_callback"

    def on_run_initialized(self, run_dir, path, **kwargs):
        print("Run initialized!")

class AnotherCallback(Callback):
    @property
    def name(self):
        return "another_callback"

    def on_run_initialized(self, run_dir, path, **kwargs):
        print("Run initialized by another callback!")

swanlab.merge_callbacks([MyCallback(), AnotherCallback()])
swanlab.init(project="my-project")

Callback Lifecycle ​

SwanLab callbacks implement the Callback base class. Common hook methods include:

  • on_run_initialized(run_dir, path, **kwargs) — Called when a run is initialized.
  • on_run_finished(state, error) — Called when a run finishes.