Skip to content

Enable/disable training on thinking tokens by configuration #117

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
mdawn65 wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Enable/disable training on thinking tokens by configuration #117

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
2 changes: 1 addition & 1 deletion skyrl-train/skyrl_train/generators/skyrl_gym_generator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ def __init__(
self.use_conversation_multi_turn = generator_cfg.use_conversation_multi_turn

# optionally use custom chat template to get loss masks (i.e. for Qwen3)
self.custom_chat_template = get_custom_chat_template(model_name)
self.custom_chat_template = get_custom_chat_template(model_name, generator_cfg.get("train_on_thinking_tokens", False))
# get generation prompt ids for the tokenizer if needed
self.generation_prompt_ids = get_generation_prompt_ids(tokenizer) if self.use_conversation_multi_turn else None
if self.skyrl_gym_cfg.max_env_workers > 0:
Expand Down
4 changes: 2 additions & 2 deletions skyrl-train/skyrl_train/generators/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,8 +28,8 @@
}


def get_custom_chat_template(model_name: str) -> str:
if "Qwen3" in model_name:
def get_custom_chat_template(model_name: str, train_on_thinking_tokens: bool = False) -> Optional[str]:
if "Qwen3" in model_name and not train_on_thinking_tokens:
return CUSTOM_CHAT_TEMPLATES["qwen3_thinking"]
else:
return None
Expand Down