Skip to content

fix bug in EvalHarnessAdapter when pipe parallel size is set to 1 #1356

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

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

fix bug in EvalHarnessAdapter when pipe parallel size is set to 1 #1356

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
14 changes: 8 additions & 6 deletions eval_tasks/eval_adapter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,8 +59,8 @@ def __init__(self, model, forward_step_fn, neox_args, batch_size=None):
self.is_main = neox_args.rank == 0
self.is_local_main = neox_args.local_rank == 0
self.is_model_parallel = neox_args.model_parallel_size > 1
self.is_pipe_parallel = self.model.is_pipe_parallel
self.is_data_parallel = self.model.is_data_parallel
self.is_pipe_parallel = getattr(self.model, 'is_pipe_parallel', False)
self.is_data_parallel = getattr(self.model, 'is_data_parallel', False)
self.is_last_stage = (
True if not self.is_pipe_parallel else model.is_last_stage()
) # only the last stage of the pipeline model will receive the logits
Expand Down Expand Up @@ -369,7 +369,11 @@ def _model_call(self, inps):
self.model.first_output_send = True
self.model.pipe_recv_buf = None

_, logits = self._forward_step_fn(model=self.model, data_iterator=inps)
_, logits = self._forward_step_fn(model=self.model, data_iterator=inps)

# since return_logits is true, forward will return 3 vals
else:
_, logits, _ = self._forward_step_fn(model=self.model, data_iterator=inps)

# gather outputs from all dp ranks:
logits = self._dp_gather(logits)
Expand All @@ -396,9 +400,7 @@ def run_eval(
):
was_training = self.model.training
self.model.eval()
in_micro_batches = (
self.model.micro_batches
) # store input microbatches - we need to set to 1 during eval, but want to return to its original value after
in_micro_batches = getattr(self.model, 'micro_batches', 1) # store input microbatches - we need to set to 1 during eval, but want to return to its original value after
self.model.micro_batches = 1
if eval_tasks is None:
eval_tasks = [
Expand Down