Skip to content

Add pytorch profiling #29

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

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

Add pytorch profiling #29

Show file tree
Hide file tree
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
4 changes: 4 additions & 0 deletions tuning/config/configs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,3 +40,7 @@ class TrainingArguments(transformers.TrainingArguments):
default=False,
metadata={"help": "Packing to be enabled in SFT Trainer, default is False"},
)
profiling: bool = field(
default=False,
metadata={"help": "Whether to enable Profiling, default is False"},
)
30 changes: 29 additions & 1 deletion tuning/sft_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,18 @@ def on_save(self, args, state, control, **kwargs):
if "pytorch_model.bin" in os.listdir(checkpoint_path):
os.remove(os.path.join(checkpoint_path, "pytorch_model.bin"))

class ProfilerCallback(TrainerCallback):
def __init__(self, prof):
self.prof = prof

def on_step_end(self, args, state, control, **kwargs):
self.prof.step()

def trace_handler(prof):
prof.export_chrome_trace(f"trace.json.gz")
prof.export_memory_timeline(f"mem.html")
prof.export_memory_timeline(f"mem.json")
prof.export_memory_timeline(f"mem.raw.json.gz")

def train(
model_args: configs.ModelArguments,
Expand Down Expand Up @@ -153,6 +165,17 @@ def train(
data_collator = DataCollatorForCompletionOnlyLM(response_template_ids, tokenizer=tokenizer, ignore_index=configs.IGNORE_INDEX)
packing = False

if train_args.profiling:
logger.info("Enabling Profiling")
profiler = torch.profiler.profile(
activities=[torch.profiler.ProfilerActivity.CPU, torch.profiler.ProfilerActivity.CUDA],
schedule=torch.profiler.schedule(wait=0, warmup=0, active=5, repeat=1),
record_shapes=True,
profile_memory=True,
with_stack=True,
on_trace_ready=trace_handler)
callbacks.append(ProfilerCallback(profiler))

trainer = SFTTrainer(
model=model,
tokenizer=tokenizer,
Expand All @@ -168,7 +191,12 @@ def train(

if run_distributed and peft_config is not None:
trainer.accelerator.state.fsdp_plugin.auto_wrap_policy = fsdp_auto_wrap_policy(model)
trainer.train()

if train_args.profiling:
with profiler as prof:
trainer.train()
else:
trainer.train()

def main(**kwargs):
parser = transformers.HfArgumentParser(dataclass_types=(configs.ModelArguments,
Expand Down