LocalLoggerCallback
Human-readable local logging: prints what's happening per step AND persists metrics + benchmark predictions to files under the run dir.
Writes \<output_dir>/\<run_key>/ : metrics.jsonl (train + val rows),
predictions/\<name>.jsonl (per benchmark), and summary.md at
run end. The "print what's happening" requirement is the per-step one-liner
(cadence print_every).
Attributes
attributenamestr= 'local_logger'attributeConfigtype= LocalLoggerConfigattributeprint_every= int(print_every)attributekeys= keysFunctions
func__init__(self, *, print_every=1, keys=None) -> Noneparamselfparamprint_everyint= 1paramkeyslist[str] | None= NoneReturns
Nonefuncon_run_start(self, ctx) -> NoneparamselfparamctxLogContextReturns
Nonefunc_write_metrics(self, step, metrics, split) -> NoneparamselfparamstepintparammetricsdictparamsplitstrReturns
Nonefuncon_step_end(self, state, step_idx, batch, metrics) -> NoneparamselfparamstateLoopStateparamstep_idxparambatchparammetricsReturns
Nonefuncon_eval(self, state, step_idx, eval_name, metrics) -> NoneparamselfparamstateLoopStateparamstep_idxparameval_nameparammetricsReturns
Nonefuncon_benchmark_eval(self, ctx, eval_result, predictions, *, step=None) -> Noneparamselfparamctxparameval_resultparampredictionsparamstep= NoneReturns
Nonefuncon_run_end(self, ctx, run_result, arm) -> Noneparamselfparamctxparamrun_resultparamarmReturns
None