AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

Multiply lr scheduler steps by num_processes. (#3983)

Browse Source 
* Multiply lr scheduler steps by `num_processes`.

* Stop multiplying steps by gradient accumulation.
This commit is contained in:
Ruoxi
2023-07-13 20:20:25 +08:00
committed by GitHub
parent 92a57a8e84
commit ece55227ff
17 changed files with 35 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
examples/controlnet/train_controlnet.py
View File

@@ -897,8 +897,8 @@ def main(args):
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
num_cycles=args.lr_num_cycles,
power=args.lr_power,
)

4
examples/custom_diffusion/train_custom_diffusion.py
View File

@@ -1007,8 +1007,8 @@ def main(args):
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/dreambooth/train_dreambooth.py
View File

@@ -1075,8 +1075,8 @@ def main(args):
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
num_cycles=args.lr_num_cycles,
power=args.lr_power,
)

4
examples/dreambooth/train_dreambooth_lora.py
View File

@@ -1039,8 +1039,8 @@ def main(args):
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
num_cycles=args.lr_num_cycles,
power=args.lr_power,
)

4
examples/instruct_pix2pix/train_instruct_pix2pix.py
View File

@@ -690,8 +690,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/research_projects/dreambooth_inpaint/train_dreambooth_inpaint.py
View File

@@ -600,8 +600,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
if args.train_text_encoder:

4
examples/research_projects/dreambooth_inpaint/train_dreambooth_inpaint_lora.py
View File

@@ -644,8 +644,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/research_projects/intel_opts/textual_inversion/textual_inversion_bf16.py
View File

@@ -481,8 +481,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(

4
examples/research_projects/intel_opts/textual_inversion_dfq/textual_inversion.py
View File

@@ -588,8 +588,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
if not train_unet:

4
examples/research_projects/lora/train_text_to_image_lora.py
View File

@@ -701,8 +701,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/research_projects/mulit_token_textual_inversion/textual_inversion.py
View File

@@ -690,8 +690,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/research_projects/multi_subject_dreambooth/train_multi_subject_dreambooth.py
View File

@@ -970,8 +970,8 @@ def main(args):
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
num_cycles=args.lr_num_cycles,
power=args.lr_power,
)

4
examples/research_projects/onnxruntime/text_to_image/train_text_to_image.py
View File

@@ -732,8 +732,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/research_projects/onnxruntime/textual_inversion/textual_inversion.py
View File

@@ -741,8 +741,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/text_to_image/train_text_to_image.py
View File

@@ -819,8 +819,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

4
examples/text_to_image/train_text_to_image_lora.py
View File

@@ -662,8 +662,8 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
)
# Prepare everything with our `accelerator`.

6
examples/textual_inversion/textual_inversion.py
View File

@@ -737,9 +737,9 @@ def main():
lr_scheduler = get_scheduler(
args.lr_scheduler,
optimizer=optimizer,
num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
num_cycles=args.lr_num_cycles * args.gradient_accumulation_steps,
num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
num_training_steps=args.max_train_steps * accelerator.num_processes,
num_cycles=args.lr_num_cycles,
)
# Prepare everything with our `accelerator`.