Skip to content

fix: gemma3 27b must now have skip_tokenizer_init=False in vllm #1721

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
yuki-97 merged 1 commit into from
Jan 7, 2026
Merged

fix: gemma3 27b must now have skip_tokenizer_init=False in vllm #1721

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
8 changes: 8 additions & 0 deletions nemo_rl/models/generation/vllm/vllm_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -388,6 +388,14 @@ def _patch_vllm_vit_flash_attn_backend():
)
# disable quantization
vllm_kwargs["hf_overrides"]["quantization_config"] = {}
elif "Gemma3ForConditionalGeneration" in getattr(
hf_config, "architectures", []
):
if self.cfg["vllm_cfg"]["skip_tokenizer_init"]:
print(
"Gemma3ForConditionalGeneration models may crash when skip_tokenizer_init is True. NeMo-RL is forcing it to False for this architecture. See https://github.com/NVIDIA-NeMo/RL/issues/1681 for more details."
)
self.cfg["vllm_cfg"]["skip_tokenizer_init"] = False

llm_kwargs = dict(
model=self.model_name,
Expand Down
Loading