LogContext

Mutable context threaded to every callback hook, in BOTH the Experiment scope (lifecycle / benchmark eval) and the TrainingLoop scope (per-step / checkpoint). Logger callbacks create the dashboard ids and write them onto :attr:ids; later hooks read them back.

The framework owns only :attr:run_key (the local output-dir name). All dashboard ids (experiment_id / group:\<name> / run_id) are callback-owned and accumulate on :attr:ids as the create_* hooks fire. Arms run sequentially, so ids["run_id"] is the current arm's.

Attributes

attributeoutput_dirPath

attributeconfig'ExperimentConfig | None'

= None

attributestoreAny

= None

Resolved store handle (EvsysStore / LocalStore / DashboardClient) or None when no store is configured.

attributerun_keystr | None

= None

attributerun_config'RunConfig | None'

= None

attributegroup_namestr | None

= None

attributeidsdict[str, str]

= field(default_factory=dict)

Callback-populated dashboard ids: experiment_id, group:\<name>, run_id. Read parents here to link children (e.g. create_run reads ids['experiment_id']).

attributeextrasdict[str, Any]

= field(default_factory=dict)

Scratch for passing values between callbacks within a run (e.g. wandb_url set by wandb_logger, read by evsys_logger's create_run).

Functions

func__init__(self, output_dir, config=None, store=None, run_key=None, run_config=None, group_name=None, ids=dict(), extras=dict()) -> None

paramself

paramoutput_dirPath

paramconfig'ExperimentConfig | None'

= None

paramstoreAny

= None

paramrun_keystr | None

= None

paramrun_config'RunConfig | None'

= None

paramgroup_namestr | None

= None

paramidsdict[str, str]

= dict()

paramextrasdict[str, Any]

= dict()

Returns

None