Skip to content

Add degenerate_group_masking option to DeepSWE training script. #1488

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
copybara-service wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −0
Open

Add degenerate_group_masking option to DeepSWE training script. #1488

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
12 changes: 12 additions & 0 deletions examples/deepswe/train_deepswe_nb.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,6 +185,15 @@
"--loss_agg_mode", type=str, default="sequence-mean-token-scale"
)
parser.add_argument("--advantage_estimator", type=str, default="rloo")
parser.add_argument(
"--degenerate_group_masking",
type=bool,
default=False,
help=(
"Whether to mask out groups whose advantages are all zero. "
"Default is False to align with rLLM DeepSWE."
),
)


# Other
Expand Down Expand Up @@ -455,6 +464,7 @@
)
LOSS_AGG_MODE = args.loss_agg_mode
ADVANTAGE_ESTIMATOR = args.advantage_estimator
DEGENERATE_GROUP_MASKING = args.degenerate_group_masking


# %%
Expand Down Expand Up @@ -778,6 +788,7 @@ def transform(entry):
"filter_statuses": FILTER_STATUSES,
"loss_agg_mode": LOSS_AGG_MODE,
"advantage_estimator": ADVANTAGE_ESTIMATOR,
"degenerate_group_masking": DEGENERATE_GROUP_MASKING,
}

grpo_config = agentic_grpo_learner.GRPOConfig(**config_kwargs)
Expand Down Expand Up @@ -856,6 +867,7 @@ def mixed_type_batch_fn(elements):
"filter_statuses": (
[s.name for s in FILTER_STATUSES] if FILTER_STATUSES else None
),
"degenerate_group_masking": DEGENERATE_GROUP_MASKING,
# Mesh topology
"num_devices": len(devices),
"rollout_mesh_fsdp": rollout_fsdp,
Expand Down
Loading