Skip to content

Fix nested vocab_size for DistillationTrainer and GOLDTrainer #5592

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
Beichen-Ma wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Fix nested vocab_size for DistillationTrainer and GOLDTrainer #5592

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
d3d3aa1
fix nested vocab_size for DistillationTrainer and GOLDTrainer
Beichen-Ma Apr 19, 2026
1f33cd4
Merge branch 'main' into fix-nested-vocab-size
Beichen-Ma Apr 22, 2026
d1e8543
Merge branch 'main' into fix-nested-vocab-size
Beichen-Ma Apr 29, 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
4 changes: 3 additions & 1 deletion tests/experimental/test_gold_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -486,7 +486,9 @@ def test_gold_trainer_init_defaults_vllm_max_model_length_to_max_length(monkeypa

class DummyStudentModel:
def __init__(self):
self.config = SimpleNamespace(_name_or_path="student", vocab_size=17)
config = SimpleNamespace(_name_or_path="student", vocab_size=17)
config.get_text_config = lambda: config
self.config = config
self.generation_config = SimpleNamespace(eos_token_id=2)
self.name_or_path = "student"

Expand Down
2 changes: 1 addition & 1 deletion trl/experimental/distillation/distillation_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -549,7 +549,7 @@ def __init__(
# ── Prepare teacher model (after super().__init__ so accelerator is ready) ──
if teacher_model is not None:
if self._local_teacher_tokenizer_matches_student:
teacher_model.resize_token_embeddings(self.model.config.vocab_size)
teacher_model.resize_token_embeddings(self.model.config.get_text_config().vocab_size)
if self.is_deepspeed_enabled:
self.teacher_model = prepare_deepspeed(teacher_model, self.accelerator)
else:
Expand Down
2 changes: 1 addition & 1 deletion trl/experimental/gold/gold_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -845,7 +845,7 @@ def __init__(
if args.disable_dropout:
disable_dropout_in_model(self.model)
if not args.use_uld_loss:
teacher_model.resize_token_embeddings(self.model.config.vocab_size)
teacher_model.resize_token_embeddings(self.model.config.get_text_config().vocab_size)

if self.is_deepspeed_enabled:
self.teacher_model = prepare_deepspeed(teacher_model, self.accelerator)
Expand Down