Skip to content
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

Jump to bottom

Load model faster for checkpoint conversion #5

Merged
merged 7 commits into from
Jul 26, 2023
Merged

Load model faster for checkpoint conversion #5

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
87b6162
add accelerate lib for faster model loading
seungahdev Jul 25, 2023
4c8f86e
formatting
seungahdev Jul 25, 2023
f23851a
remove torch
seungahdev Jul 26, 2023
8751c8f
fix load option w/o init_empty_weight context
seungahdev Jul 26, 2023
5bbf7aa
formatting
seungahdev Jul 26, 2023
860b28e
fix accelerate lib to mllib
seungahdev Jul 26, 2023
6b2acef
add einops for mllib
seungahdev Jul 26, 2023
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
15 changes: 9 additions & 6 deletions periflow/sdk/resource/checkpoint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -822,6 +822,7 @@ def convert(

"""
try:
import accelerate # type: ignore[import] # pylint: disable=import-outside-toplevel
seungahdev marked this conversation as resolved.
Show resolved Hide resolved
import torch # type: ignore[import] # pylint: disable=import-outside-toplevel
except ImportError as exc:
raise CheckpointConversionError(
Expand Down Expand Up @@ -900,12 +901,14 @@ def convert(
"Start loading Hugging Face checkpoint(%s) for conversion...",
model_name_or_path,
)
state_dict = hf_factory.from_pretrained(
model_name_or_path,
torch_dtype=torch.float32,
cache_dir=cache_dir,
trust_remote_code=True,
).state_dict()
with accelerate.init_empty_weights():
seungahdev marked this conversation as resolved.
Show resolved Hide resolved
state_dict = hf_factory.from_pretrained(
model_name_or_path,
torch_dtype=torch.float32,
cache_dir=cache_dir,
trust_remote_code=True,
device_map="cpu",
).state_dict()
logger.info(
"Hugging Face checkpoint(%s) is successfully loaded!",
model_name_or_path,
Expand Down
Loading