From 3d2f33df07978912a0c491965768fec6d58a14aa Mon Sep 17 00:00:00 2001 From: Stephen Roller Date: Wed, 21 Dec 2022 11:51:17 -0800 Subject: [PATCH] Fix config.yml dump in training runs. --- metaseq/cli/train.py | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/metaseq/cli/train.py b/metaseq/cli/train.py index f23371d32..ef85cbb85 100644 --- a/metaseq/cli/train.py +++ b/metaseq/cli/train.py @@ -67,11 +67,15 @@ def main(cfg: DictConfig) -> None: checkpoint_utils.verify_checkpoint_directory(cfg.checkpoint.save_dir) - if distributed_utils.is_master(cfg.distributed_training): + if distributed_utils.is_master(cfg.distributed_training) and os.environ.get( + "METASEQ_SAVE_DIR" + ): # save a (vaguely human readable) copy of the training config + # TODO(roller): only works when launched with a sweep script + # should fix that OmegaConf.save( config=_flatten_config(cfg), - f=os.path.join(cfg.checkpoint.save_dir, "config.yml"), + f=os.path.join(os.environ["METASEQ_SAVE_DIR"], "config.yml"), ) if (