Skip to content

feature: add kl divergence with ref model #5

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

Merged
LovelyBuggies merged 2 commits into from
Jun 30, 2026
Merged

feature: add kl divergence with ref model #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
49db8a3
add kl
LovelyBuggies Jun 30, 2026
fe228f5
allow ref on separate devices
LovelyBuggies Jun 30, 2026
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
3 changes: 3 additions & 0 deletions train_ac.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -301,6 +301,9 @@ def main() -> None:
eval_num_samples=ac_cfg.get("eval_num_samples", 4),
eval_batch_size=ac_cfg.get("eval_batch_size", 1),
logging_steps=ac_cfg.get("logging_steps", 1),
reference_kl_enabled=ac_cfg.get("reference_kl_enabled", False),
reference_kl_coef=ac_cfg.get("reference_kl_coef", 0.1),
reference_devices=ac_cfg.get("reference_devices", None),
),
train_dataset=train_dataset,
eval_dataset=eval_dataset,
Expand Down
3 changes: 3 additions & 0 deletions train_grpo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -235,6 +235,9 @@ def main():
eval_batch_size=grpo_cfg.get("eval_batch_size", 1),
train_batch_size=grpo_cfg.get("train_batch_size"),
advantage_normalization=grpo_cfg.get("advantage_normalization", True),
reference_kl_enabled=grpo_cfg.get("reference_kl_enabled", False),
reference_kl_coef=grpo_cfg.get("reference_kl_coef", 0.1),
reference_devices=grpo_cfg.get("reference_devices", None),
)

import rewards.arxiv_rewards as arxiv_rewards
Expand Down
3 changes: 3 additions & 0 deletions train_iac.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -328,6 +328,9 @@ def main() -> None:
eval_num_samples=iac_cfg.get("eval_num_samples", 4),
eval_batch_size=iac_cfg.get("eval_batch_size", 1),
logging_steps=iac_cfg.get("logging_steps", 50),
reference_kl_enabled=iac_cfg.get("reference_kl_enabled", False),
reference_kl_coef=iac_cfg.get("reference_kl_coef", 0.1),
reference_devices=iac_cfg.get("reference_devices", None),
),
train_dataset=train_dataset,
eval_dataset=eval_dataset,
Expand Down
3 changes: 3 additions & 0 deletions train_maac.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -323,6 +323,9 @@ def main() -> None:
eval_num_samples=maac_cfg.get("eval_num_samples", 4),
eval_batch_size=maac_cfg.get("eval_batch_size", 1),
logging_steps=maac_cfg.get("logging_steps", 50),
reference_kl_enabled=maac_cfg.get("reference_kl_enabled", False),
reference_kl_coef=maac_cfg.get("reference_kl_coef", 0.1),
reference_devices=maac_cfg.get("reference_devices", None),
),
train_dataset=train_dataset,
eval_dataset=eval_dataset,
Expand Down
3 changes: 3 additions & 0 deletions train_magrpo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -332,6 +332,9 @@ def main():
eval_interval=magrpo_cfg.get("eval_interval", 20),
eval_num_samples=magrpo_cfg.get("eval_num_samples", 4),
eval_batch_size=magrpo_cfg.get("eval_batch_size", 1),
reference_kl_enabled=magrpo_cfg.get("reference_kl_enabled", False),
reference_kl_coef=magrpo_cfg.get("reference_kl_coef", 0.1),
reference_devices=magrpo_cfg.get("reference_devices", None),
)

import rewards.arxiv_rewards as arxiv_rewards
Expand Down