SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Load state in else (#25318)

Browse Source 
* Load else

* New approach

* Propagate
This commit is contained in:
Zach Mueller
2023-08-08 05:41:00 -04:00
committed by GitHub
parent 36d5b8b06c
commit 01ab39b65f
12 changed files with 72 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
examples/pytorch/image-classification/run_image_classification_no_trainer.py
View File

@@ -454,14 +454,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/image-pretraining/run_mim_no_trainer.py
View File

@@ -664,14 +664,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/language-modeling/run_clm_no_trainer.py
View File

@@ -575,14 +575,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/language-modeling/run_mlm_no_trainer.py
View File

@@ -613,14 +613,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/multiple-choice/run_swag_no_trainer.py
View File

@@ -560,14 +560,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py
View File

@@ -797,14 +797,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/question-answering/run_qa_no_trainer.py
View File

@@ -828,14 +828,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py
View File

@@ -558,14 +558,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/summarization/run_summarization_no_trainer.py
View File

@@ -629,14 +629,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/text-classification/run_glue_no_trainer.py
View File

@@ -507,14 +507,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/token-classification/run_ner_no_trainer.py
View File

@@ -667,14 +667,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]

8
examples/pytorch/translation/run_translation_no_trainer.py
View File

@@ -610,14 +610,18 @@ def main():
# Potentially load in the weights and states from a previous save
if args.resume_from_checkpoint:
if args.resume_from_checkpoint is not None or args.resume_from_checkpoint != "":
accelerator.print(f"Resumed from checkpoint: {args.resume_from_checkpoint}")
accelerator.load_state(args.resume_from_checkpoint)
checkpoint_path = args.resume_from_checkpoint
path = os.path.basename(args.resume_from_checkpoint)
else:
# Get the most recent checkpoint
dirs = [f.name for f in os.scandir(os.getcwd()) if f.is_dir()]
dirs.sort(key=os.path.getctime)
path = dirs[-1] # Sorts folders by date modified, most recent checkpoint is the last
checkpoint_path = path
path = os.path.basename(checkpoint_path)
accelerator.print(f"Resumed from checkpoint: {checkpoint_path}")
accelerator.load_state(path)
# Extract `epoch_{i}` or `step_{i}`
training_difference = os.path.splitext(path)[0]