Skip to content
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.

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

fix PPO #2377

Merged
merged 7 commits into from
Nov 11, 2024
Merged

fix PPO #2377

merged 7 commits into from
Nov 11, 2024

Conversation

hjh0119
Copy link
Collaborator

@hjh0119 hjh0119 commented Nov 4, 2024

PR type

  • Bug Fix
  • New Feature
  • Document Updates
  • More Models or Datasets Support

PR information

related issue:
#2316
#2296
#2330
#2285

Experiment results

Paste your experiment result here(if needed).

@Jintao-Huang
Copy link
Collaborator

Is it still in WIP status, or can it be reviewed now

@hjh0119
Copy link
Collaborator Author

hjh0119 commented Nov 5, 2024

Is it still in WIP status, or can it be reviewed now

It's still under testing now

@hjh0119 hjh0119 changed the title [WIP] fix PPO fix PPO Nov 10, 2024
Comment on lines +1798 to +1824
from contextlib import nullcontext, contextmanager

@contextmanager
def ppocontext():
from transformers.integrations.deepspeed import HfTrainerDeepSpeedConfig
from transformers.utils import is_sagemaker_mp_enabled
if is_sagemaker_mp_enabled():
import smdistributed.modelparallel.torch as smp
smp.init()
old_trainer_config_process = HfTrainerDeepSpeedConfig.trainer_config_process

def trainer_config_process(self, args, auto_find_batch_size=False):
if args.world_size is None:
if args.distributed_state is not None:
return args.distributed_state.num_processes
elif is_sagemaker_mp_enabled():
return smp.dp_size() if not smp.state.cfg.prescaled_batch else smp.rdp_size()
return 1
return old_trainer_config_process(self, args, auto_find_batch_size)

HfTrainerDeepSpeedConfig.trainer_config_process = trainer_config_process
yield
HfTrainerDeepSpeedConfig.trainer_config_process = old_trainer_config_process

context = nullcontext if self.rlhf_type != 'ppo' else ppocontext
with context():
super().__post_init__()
Copy link
Collaborator Author

Choose a reason for hiding this comment

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

issue #2296

@Jintao-Huang Jintao-Huang merged commit 1e878b9 into modelscope:main Nov 11, 2024
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants