Skip to content
New issue

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

#

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

Already on GitHub? # to your account

fix: unscale fp16 gradient problem & potential error (#6086) #6231

Merged
merged 3 commits into from
Dec 21, 2023
Merged

fix: unscale fp16 gradient problem & potential error (#6086) #6231

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1b4ad5d
fix: unscale fp16 gradient problem & potential error (#6086)
lvzii Dec 19, 2023
b954499
Merge branch 'main' into fix-sdxl-train-script
sayakpaul Dec 20, 2023
ab3b41a
Merge branch 'main' into fix-sdxl-train-script
sayakpaul Dec 20, 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
14 changes: 14 additions & 0 deletions examples/text_to_image/train_text_to_image_lora_sdxl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -640,6 +640,17 @@ def main(args):
text_encoder_one.add_adapter(text_lora_config)
text_encoder_two.add_adapter(text_lora_config)

# Make sure the trainable params are in float32.
if args.mixed_precision == "fp16":
models = [unet]
if args.train_text_encoder:
models.extend([text_encoder_one, text_encoder_two])
for model in models:
for param in model.parameters():
# only upcast trainable parameters (LoRA) into fp32
if param.requires_grad:
param.data = param.to(torch.float32)
Comment on lines +643 to +652
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Works for me!


# create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
def save_model_hook(models, weights, output_dir):
if accelerator.is_main_process:
Expand Down Expand Up @@ -1187,6 +1198,9 @@ def compute_time_ids(original_size, crops_coords_top_left):
torch.cuda.empty_cache()

# Final inference
# Make sure vae.dtype is consistent with the unet.dtype
if args.mixed_precision == "fp16":
vae.to(weight_dtype)
Comment on lines +1201 to +1203
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is not needed in my opinion. We already set the torch_dtype in the pipeline during loading it.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If no other pretrained_vae_model_name_or_path is set, then the vae is set to float32,

if args.pretrained_vae_model_name_or_path is None:
        vae.to(accelerator.device, dtype=torch.float32)

and the pipeline here does not reload vae. So vae.dtype float32 != unet.dtype fp16, which in my tests causes RuntimeError: Input type (c10::Half) and bias type (float) should be the same

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah okay got it!

# Load previous pipeline
pipeline = StableDiffusionXLPipeline.from_pretrained(
args.pretrained_model_name_or_path,
Expand Down