Skip to content

FIX FSDP plugin update for QLoRA #36720

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ArthurZucker merged 2 commits into from
Mar 21, 2025
Merged

FIX FSDP plugin update for QLoRA #36720

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f86a3fd
FIX FSDP plugin update for QLoRA
BenjaminBossan Mar 14, 2025
2fca776
Merge branch 'main' into fix-fsdp-qlora-plugin-update
SunMarc Mar 20, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 4 additions & 5 deletions src/transformers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5202,18 +5202,17 @@ def propagate_args_to_deepspeed(self, auto_find_batch_size=False):

def _fsdp_qlora_plugin_updates(self):
if self.is_fsdp_enabled and _is_peft_model(self.model):
from peft import LoraConfig
from peft import PeftConfig
from peft.utils.other import fsdp_auto_wrap_policy

if isinstance(self.model.active_peft_config, LoraConfig):
fsdp_plugin = self.accelerator.state.fsdp_plugin
fsdp_plugin.auto_wrap_policy = fsdp_auto_wrap_policy(self.model)
if isinstance(self.model.active_peft_config, PeftConfig):
self.accelerator.state.fsdp_plugin.auto_wrap_policy = fsdp_auto_wrap_policy(self.model)
if (
getattr(self.model, "quantization_method", None) == QuantizationMethod.BITS_AND_BYTES
and self.model.hf_quantizer.quantization_config.bnb_4bit_quant_storage.is_floating_point
and version.parse(accelerate_version) > version.parse("0.27.0")
):
fsdp_plugin.set_mixed_precision(
self.accelerator.state.fsdp_plugin.set_mixed_precision(
self.model.hf_quantizer.quantization_config.bnb_4bit_quant_storage, override=True
)

Expand Down
Loading