SIGN IN SIGN UP

Pretrain, finetune ANY AI model of ANY size on 1 or 10,000+ GPUs with zero code changes.

0 0 104 Python

Expose `weights_only` for loading checkpoints with `Trainer`, `LightningModule`, `LightningDataModule` (#21072)

* change weights_only default to True

* add docs on weights_only arg

* add weights_only arg to checkpoint save. weights_only during test set based on ckpt version

* add weights_only arg to checkpoint_io

* woops, reverting changes

* permissions too

* fix link

* fix another link

* datamodule weights_only args

* wip: try safe_globals context manager for tests

* add weights_only arg to _run_standard_hparams_test

* weights_only=False when adding extra_args

* switch to lightning_utilities.cli requirements set-oldest (#21077)

* bump: try `deepspeed >=0.14.1,<=0.15.0` (#21076)

* try `deepspeed >=0.14.1,<=0.15.0`
* drop from oldest
* pip uninstall -y deepspeed
* error::DeprecationWarning

* weights_only=True default for torch>=2.6

* changelog

* ignore torch.load futurewarning

* add .*

* will this woork

* weights_only according pl version

* set env var TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 for pl < 1.5.0

* weights_only=False for omegaconf hparams test

* default to weights_only=true for loading from state_dict from url

* weights_only=False for hydra

* Update src/lightning/fabric/utilities/cloud_io.py

Co-authored-by: Jirka Borovec <6035284+Borda@users.noreply.github.com>

* defaults for weights_only in torch.hub.load_state_dict_from_url

* default to weights_only=False for torch.hub.load_state_dict_from_url

* add weights_only to trainer.fit, validate, test, predict

* fix tests

* add weights_only arg

* specify weights_only kwarg

* weights_only for fsdp load

* Apply suggestions from code review

* Apply suggestions from code review

* default is none

* add weights_only args to strategies

* trainer default to weights_only=None

* wip: fix typing dump_checkpoint

* Apply suggestions from code review

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* weights_only as last arg

* asset called with none

* weights_only=False for torch>=2.6 in tests

* fix changelog description

* Empty-Commit

* Empty-Commit

* trigger ci

* skip ddp_fork on macos

---------

Co-authored-by: Nicki Skafte Detlefsen <skaftenicki@gmail.com>
Co-authored-by: Jirka Borovec <6035284+Borda@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: thomas chaton <thomas@grid.ai>
Co-authored-by: Jirka B <j.borovec+github@gmail.com>
Co-authored-by: jirka <jirka.borovec@seznam.cz>
Co-authored-by: Deependu <deependujha21@gmail.com>
S
Shion Matsumoto committed
29abe6e47deefc628b7ede09601c05838fcf140b
Parent: b82db78
Committed by GitHub <noreply@github.com> on 11/7/2025, 6:00:46 AM