Skip to content

Add callbacks to finetune function #2278

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
michal2409 merged 2 commits into from
Feb 9, 2026
Merged

Add callbacks to finetune function #2278

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
559730a
Add callbacks to finetune function
michal2409 Feb 9, 2026
cdbb03e
fix tests
michal2409 Feb 9, 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
5 changes: 4 additions & 1 deletion src/megatron/bridge/training/finetune.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.

from megatron.bridge.training.callbacks import Callback, CallbackManager
from megatron.bridge.training.config import ConfigContainer
from megatron.bridge.training.forward_step_func_types import ForwardStepCallable
from megatron.bridge.training.pretrain import pretrain
Expand All @@ -22,6 +23,7 @@
def finetune(
config: ConfigContainer,
forward_step_func: ForwardStepCallable,
callbacks: list[Callback] | CallbackManager | None = None,
) -> None:
"""Main function to run the finetuning.

Expand All @@ -34,6 +36,7 @@ def finetune(
- 3 args: (data_iterator, model, return_schedule_plan=False)
OR (state: GlobalState, data_iterator, model)
- 4 args: (state: GlobalState, data_iterator, model, return_schedule_plan=False)
callbacks: Optional list of Callback instances, a CallbackManager, or None.

Note:
Use the signature with GlobalState type hint for full access to configuration, timers, and training state.
Expand All @@ -47,4 +50,4 @@ def finetune(
assert config.checkpoint.pretrained_checkpoint is not None or config.checkpoint.load is not None, (
"Finetuning requires a loading from a pretrained checkpoint or resuming from a checkpoint"
)
return pretrain(config, forward_step_func)
return pretrain(config, forward_step_func, callbacks=callbacks)
Loading