Skip to content

Fix lora merge script #8113

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
cuichenx merged 2 commits into from
Jan 3, 2024
Merged

Fix lora merge script #8113

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f6cd3d7
fix lora merge script
cuichenx Jan 2, 2024
46902f0
Merge branch 'main' into chcui/fix_lora_merge
arendu Jan 3, 2024
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
12 changes: 5 additions & 7 deletions scripts/nlp_language_modeling/merge_lora_weights/merge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,6 @@
from torch.utils.data import DataLoader, Dataset

from nemo.collections.nlp.models.language_modeling.megatron_gpt_model import MegatronGPTModel
from nemo.collections.nlp.models.language_modeling.megatron_gpt_peft_models import MegatronGPTLoRAModel
from nemo.collections.nlp.models.language_modeling.megatron_gpt_sft_model import MegatronGPTSFTModel
from nemo.collections.nlp.modules.common.megatron.megatron_init import fake_initialize_model_parallel
from nemo.collections.nlp.parts.nlp_overrides import NLPDDPStrategy, NLPSaveRestoreConnector
Expand Down Expand Up @@ -72,7 +71,9 @@ def load_lora(lora_nemo, tp):

l = torch.load(ckpt_file, map_location=torch.device('cpu'))
lora_state_dict[i] = l
return lora_state_dict
config_file = f"{tmpdir}/model_config.yaml"
lora_config = OmegaConf.load(config_file)
return lora_state_dict, lora_config


def fix_for_O2(state_dict):
Expand Down Expand Up @@ -195,12 +196,8 @@ def main(cfg) -> None:
else:
raise ValueError("need at least a nemo file or checkpoint dir")

lora_model_cfg = MegatronGPTLoRAModel.restore_from(
restore_path=cfg.lora_model_path, trainer=trainer, return_config=True, mcore=model.mcore_gpt,
)

# load the lora weights on cpu for all ranks of the lora model
lora_weights = load_lora(cfg.lora_model_path, model.cfg.tensor_model_parallel_size)
lora_weights, lora_model_cfg = load_lora(cfg.lora_model_path, model.cfg.tensor_model_parallel_size)

# merge the lora weights with the base model, for this current rank.
merged_weights = merge(
Expand All @@ -209,6 +206,7 @@ def main(cfg) -> None:
tp=model.cfg.tensor_model_parallel_size,
num_layers=model.cfg.num_layers,
curr_rank=model.global_rank,
mcore=model.mcore_gpt,
)

# load the merged_weights back into the base model, for this current rank.
Expand Down