From 968ac091c0c2fc628baed22215f5d295772341c5 Mon Sep 17 00:00:00 2001 From: ananthsub Date: Thu, 8 Apr 2021 00:26:59 -0700 Subject: [PATCH] Remove hardcoding of rank_zero_only.rank in accelerator connector (#6878) --- CHANGELOG.md | 4 +++- .../trainer/connectors/accelerator_connector.py | 7 ------- 2 files changed, 3 insertions(+), 8 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index c05b5f2e28..f79a5b3549 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -177,7 +177,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/). ### Fixed -- Set better defaults for `rank_zero_only.rank` when training is launched with SLURM and torchelastic ([#6802](https://github.com/PyTorchLightning/pytorch-lightning/pull/6802/)) +- Set better defaults for `rank_zero_only.rank` when training is launched with SLURM and torchelastic: + * Support SLURM and torchelastic global rank environment variables ([#5715](https://github.com/PyTorchLightning/pytorch-lightning/pull/5715)) + * Remove hardcoding of local rank in accelerator connector ([#6878](https://github.com/PyTorchLightning/pytorch-lightning/pull/6878)) - Made the `Plugin.reduce` method more consistent across all Plugins to reflect a mean-reduction by default ([#6011](https://github.com/PyTorchLightning/pytorch-lightning/pull/6011)) diff --git a/pytorch_lightning/trainer/connectors/accelerator_connector.py b/pytorch_lightning/trainer/connectors/accelerator_connector.py index 1e00d33cdf..aa52ec1c40 100644 --- a/pytorch_lightning/trainer/connectors/accelerator_connector.py +++ b/pytorch_lightning/trainer/connectors/accelerator_connector.py @@ -59,7 +59,6 @@ from pytorch_lightning.utilities import ( device_parser, DeviceType, DistributedType, - rank_zero_only, ) from pytorch_lightning.utilities.distributed import rank_zero_info, rank_zero_warn from pytorch_lightning.utilities.exceptions import MisconfigurationException @@ -112,12 +111,6 @@ class AcceleratorConnector(object): self._training_type_plugin: Optional[TrainingTypePlugin] = None self._cluster_environment: Optional[ClusterEnvironment] = None - # init the default rank if exists - # we need to call this here or NVIDIA flags and other messaging in init will show on all ranks - # this way we only show it on rank 0 - if "LOCAL_RANK" in os.environ: - rank_zero_only.rank = int(os.environ["LOCAL_RANK"]) - # for gpus allow int, string and gpu list if auto_select_gpus and isinstance(gpus, int): self.gpus = pick_multiple_gpus(gpus)