Skip to content

fix: handle gemma 3 weights prefix during hf conversion #2156

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
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix: handle gemma 3 weights prefix during hf conversion #2156

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
f4826ef
fix: handle gemma 3 weights prefix during hf conversion
scascar Nov 11, 2025
b55b7ad
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Nov 11, 2025
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
26 changes: 16 additions & 10 deletions litgpt/scripts/convert_hf_checkpoint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@

from litgpt.config import Config
from litgpt.utils import (
_TRANSFORMERS_GREATER_EQUAL_4_52,
extend_checkpoint_dir,
incremental_save,
lazy_load,
Expand Down Expand Up @@ -292,15 +291,6 @@ def copy_weights_gemma_2(
pbar.update(progress_per_file)


GEMMA3_LANGUAGE_MODEL_PREFIX = "model.language_model" if _TRANSFORMERS_GREATER_EQUAL_4_52 else "language_model.model"

GEMMA3_VISION_MODEL_PREFIX = "model.vision_tower" if _TRANSFORMERS_GREATER_EQUAL_4_52 else "vision_tower"

GEMMA3_MM_PROJECTOR_PREFIX = (
"model.multi_modal_projector" if _TRANSFORMERS_GREATER_EQUAL_4_52 else "multi_modal_projector"
)


def copy_weights_gemma_3(
qkv_weights: Dict[int, List[Optional[NotYetLoadedTensor]]],
state_dict: Dict[str, torch.Tensor],
Expand All @@ -312,6 +302,22 @@ def copy_weights_gemma_3(
debug_mode: Optional[bool] = False,
config: Optional[Config] = None,
) -> None:
GEMMA3_LANGUAGE_MODEL_PREFIX = (
"model.language_model"
if any(k.startswith("model.language_model") for k in hf_weights)
else "language_model.model"
)

GEMMA3_VISION_MODEL_PREFIX = (
"model.vision_tower" if any(k.startswith("model.vision_tower") for k in hf_weights) else "vision_tower"
)

GEMMA3_MM_PROJECTOR_PREFIX = (
"model.multi_modal_projector"
if any(k.startswith("model.multi_modal_projector") for k in hf_weights)
else "multi_modal_projector"
)

weight_map = {
"model.embed_tokens.weight": "transformer.wte.weight",
"model.layers.{}.self_attn.q_proj.weight": None,
Expand Down
1 change: 0 additions & 1 deletion litgpt/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,6 @@
_LITDATA_AVAILABLE = RequirementCache("litdata")
_LITSERVE_AVAILABLE = RequirementCache("litserve")
_JINJA2_AVAILABLE = RequirementCache("jinja2")
_TRANSFORMERS_GREATER_EQUAL_4_52 = RequirementCache("transformers>=4.52.0")
_SAFETENSORS_AVAILABLE = RequirementCache("safetensors")
_HF_TRANSFER_AVAILABLE = RequirementCache("hf_transfer")

Expand Down