Skip to content

[Intel GPU] enable use of dinov2 models for offload benchmark_low_bit_adam #3191

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[Intel GPU] enable use of dinov2 models for offload benchmark_low_bit_adam #3191

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
18 changes: 14 additions & 4 deletions benchmarks/benchmark_low_bit_adam.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,11 +120,18 @@ def get_parser():

def get_dloader(args, training: bool):
transforms = [v2.ToImage()]
input_size = (
args.data_config["input_size"][1] if args.data_config is not None else 224
) # Standard ViT input size

if training:
transforms.extend([v2.RandomResizedCrop(224), v2.RandomHorizontalFlip()])
transforms.extend([v2.RandomResizedCrop(input_size), v2.RandomHorizontalFlip()])
else:
transforms.extend([v2.Resize(256), v2.CenterCrop(224)])
# For validation, resize to slightly larger then center crop
if "dinov2" in args.model.lower():
input_size = 518 # DINOv2 models expect 518x518
resize_size = int(input_size * 256 / 224) # Scale proportionally (584 for 518)
transforms.extend([v2.Resize(resize_size), v2.CenterCrop(input_size)])

transforms.append(v2.ToDtype(torch.float32, scale=True))
transforms.append(
Expand Down Expand Up @@ -207,12 +214,15 @@ def evaluate_model(model, args):
dir="/tmp",
mode="disabled" if args.project is None else None,
)
dloader = get_dloader(args, True)
print(f"Train dataset: {len(dloader.dataset):,} images")

model = timm.create_model(
args.model, pretrained=True, num_classes=45, **args.model_kwargs
)
args.data_config = timm.data.resolve_model_data_config(model)

dloader = get_dloader(args, True)
print(f"Train dataset: {len(dloader.dataset):,} images")

if args.checkpoint_activations:
model.set_grad_checkpointing()
if args.full_bf16:
Expand Down
Loading