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

Fix "load_model_sharded" when "--save_optimizer" enabled #128

Closed
wants to merge 1 commit into from
Closed

Fix "load_model_sharded" when "--save_optimizer" enabled #128

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
19 changes: 16 additions & 3 deletions model_checkpointing/checkpoint_handler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,8 @@
from datetime import datetime
import torch
import time
import torch.optim as optim
from configs import fsdp_config, train_config

from torch.distributed.fsdp import (
FullyShardedDataParallel as FSDP,
Expand Down Expand Up @@ -63,9 +65,18 @@ def load_model_sharded(model, rank, cfg):
if rank == 0:
print(f"loading model from model path: {load_dir} ")
reader = FileSystemReader(load_dir)

with FSDP.state_dict_type(model, StateDictType.SHARDED_STATE_DICT):
checkpoint = {"model": model.state_dict()}
# if user specified save optimizer - initialize key in dict
if train_config.save_optimizer:
optimizer = optim.AdamW(
model.parameters(),
lr=train_config.lr,
weight_decay=0.0,
)
checkpoint = {"model": model.state_dict(), "optim": FSDP.optim_state_dict(model, optimizer)}
else:
checkpoint = {"model": model.state_dict()}
if rank == 0:
ck = checkpoint.keys()
print(f" checkpoint key len = {len(ck)} and \n keys = {ck}")
Expand All @@ -79,6 +90,8 @@ def load_model_sharded(model, rank, cfg):
ck = checkpoint.keys()
print(f" checkpoint key len = {len(ck)} and \n keys = {ck}")
model.load_state_dict(checkpoint["model"])
if train_config.save_optimizer:
optimizer.load_state_dict(checkpoint["optim"])
if rank == 0:
print(f"Sharded state checkpoint loaded from {load_dir}")

Expand Down Expand Up @@ -264,4 +277,4 @@ def load_sharded_model_single_gpu(model,model_path):
model.load_state_dict(state_dict["model"])

print(f"Sharded state checkpoint loaded from {model_path}")
return model
return model
Loading