From 403c0714d1dd019a481022afce5df75a9963ecd9 Mon Sep 17 00:00:00 2001 From: Stas Bekman Date: Thu, 28 Dec 2023 10:29:26 -0800 Subject: [PATCH] Update dataclasses.py (#2292) --- src/accelerate/utils/dataclasses.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/accelerate/utils/dataclasses.py b/src/accelerate/utils/dataclasses.py index 9da98de72a2..1008521a2bb 100644 --- a/src/accelerate/utils/dataclasses.py +++ b/src/accelerate/utils/dataclasses.py @@ -930,10 +930,10 @@ class FullyShardedDataParallelPlugin: use_orig_params: bool = field( default=True, metadata={ - "help": "If `True`, allows non-uniform `requires_grad` during init, which means support for interspersed frozen and trainable paramteres. " + "help": "If `True`, allows non-uniform `requires_grad` during init, which means support for interspersed frozen and trainable parameters. " "Useful in cases such as parameter-efficient fine-tuning. " "Please refer this [blog](https://dev-discuss.pytorch.org/t/rethinking-pytorch-fully-sharded-data-parallel-fsdp-from-first-principles/1019). " - "This also enables to have different optimizer param groups. This should be `True` when creating optimizer object before preparing/wrapping the model with FSDP." + "This also enables multiple optimizer param groups. This should be `True` when creating an optimizer object before preparing/wrapping the model with FSDP." }, ) param_init_fn: Optional[Callable[[torch.nn.Module], None]] = field(