maxent_grpo.training.runtime.logging

Logging utilities (primarily W&B) for the training stack.

Functions

_ensure_wandb_installed()	Best-effort runtime install of wandb when it's missing.
_env_rank_zero()	Return True when common environment hints describe the primary node.
_git_sha()	Return the short git SHA for the current repo if available.
_is_primary_wandb_process(accelerator)	Return True when the current process should initialize a W&B run.
_log_wandb(run, metrics, step)	Safely log metrics to a W&B run.
_maybe_init_wandb_run(accelerator, ...)	Initialize a W&B run when report_to includes wandb.
_report_to_contains(report_to, target)	Case-insensitive membership check for TrainingArguments.report_to.
_torch_rank_zero()	Return True when a torch distributed rank 0 process is running.
_wandb_error_types()	Return exception types that should be suppressed during W&B logging.
log_run_header([training_args])	Log a consistent run header with recipe and resolved method identity.
resolve_run_metadata([training_args])	Return run-level metadata for logging consistency.

maxent_grpo.training.runtime.logging.log_run_header(training_args=None)

Log a consistent run header with recipe and resolved method identity.

Parameters:

training_args (Any | None) – Optional training config used to resolve metadata.

Returns:

Metadata dictionary emitted to the logs.

Return type:

dict[str, str]

maxent_grpo.training.runtime.logging.resolve_run_metadata(training_args=None)

Return run-level metadata for logging consistency.

Parameters:

training_args (Any | None) – Optional training config used to read recipe/method fields.

Returns:

Mapping with git SHA, recipe path, and resolved method identity.

Return type:

dict[str, str]