CLI Reference¶

Usage¶

pretrain-experiments <config.yaml> [options]

Or equivalently:

python -m pretrain_experiments <config.yaml> [options]

Flags¶

Flag	Description
`--resume_run_id <id>`	Resume a previous W&B run by its run ID. Also use this to add new evaluations to an existing run.
`--add-step-to-run-name`	Append the checkpoint step number to the W&B run name.
`--delete-experiment-folder`	Delete the experiment output folder before starting.
`--dry-run`	Process configs and print commands without running training or evaluation scripts.

Config overrides¶

Any config parameter can be overridden from the command line using dot notation. The override value replaces the corresponding key in the parsed YAML config.

pretrain-experiments config.yaml --training.num_steps 100
pretrain-experiments config.yaml --wandb.name my-run
pretrain-experiments config.yaml --model.checkpoint_step 5000

Multiple overrides can be combined:

pretrain-experiments config.yaml --training.num_steps 50 --wandb.name short-run

Environment variables¶

These variables are set automatically at startup and can be used in config files via ${VAR_NAME}:

Variable	Description
`PRETRAIN_EXPERIMENTS`	Root directory of the pretrain-experiments repository
`OLMO_REPO`	Root of the OLMo repository (if `olmo` is installed)
`OLMO_CORE_REPO`	Root of the OLMo-Core repository (if `olmo_core` is installed)

These variables must be set by the user as needed:

Variable	Description	Default
`EXPERIMENTS_SAVE_PATH`	Base directory for experiment outputs	`/weka/luxburg/sbordt10/single_training_run/`
`OLMO_PRIVATE_PATH`	Path to OLMo-Private repository	`/weka/luxburg/sbordt10/OLMo-Private`
`OLMES_EXECUTABLE`	Path to the `olmes` binary (for OLMES evaluations)	—
`INFERENCE_DEFAULTS_PATH`	Path to inference engine defaults YAML file	—