Skip to content

Move some conversion mappings to PrefixChange #45567

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
Cyrilvallez merged 1 commit into from
Apr 22, 2026
+3 −18
Merged

Move some conversion mappings to PrefixChange #45567

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
21 changes: 3 additions & 18 deletions src/transformers/conversion_mapping.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -151,19 +151,10 @@ def _build_checkpoint_conversion_mapping():
WeightRenaming(source_patterns=r"^visual", target_patterns="model.visual"),
],
"colqwen2": [
WeightRenaming(source_patterns=r"vlm.model", target_patterns="vlm"),
PrefixChange(prefix_to_remove="model", model_prefix="vlm"),
WeightRenaming(source_patterns=r"vlm(?!\.(language_model|visual))", target_patterns="vlm.language_model"),
],
"timm_wrapper": [
# Simply add the prefix `timm_model`. Similar to `base_model_prefix` but also removes prefix
# when saving. TODO: Would be probably much cleaner with a `add_prefix` argument in WeightRenaming
# Note: we don't add `timm_model` when it is part of a bigger VLM, because they already have `timm_model`
# saved in state dict keys. Thus the look behind check. Should be fixed by proper `add_prefix`!
WeightRenaming(
source_patterns=r"^(?!(?:model\.|backbone\.|tower\.))(.+)$",
target_patterns=r"timm_model.\1",
)
],
"timm_wrapper": [PrefixChange(prefix_to_add="timm_model")],
"pi0": [
WeightRenaming(source_patterns=r"state_proj", target_patterns="embed_action_time.state_proj"),
WeightRenaming(source_patterns=r"action_in_proj", target_patterns="embed_action_time.action_in_proj"),
Expand Down Expand Up @@ -201,13 +192,7 @@ def _build_checkpoint_conversion_mapping():
WeightRenaming("attention_layer_norm", "input_layernorm"),
WeightRenaming("feedforward_layer_norm", "post_attention_layernorm"),
],
"qwen3_5_text": [
# Note: the lookbehind on the target is to avoid replacing bigger matches when the model is a submodel of
# the ForConditionalGeneration model
WeightRenaming(
source_patterns=r"^model.language_model.", target_patterns=r"^model.(?!(?:language_model.|visual.))"
),
],
"qwen3_5_text": [PrefixChange(prefix_to_remove="language_model", model_prefix="model")],
"sam3_tracker": [
WeightRenaming(
source_patterns=r"detector_model.vision_encoder.backbone.", target_patterns="vision_encoder.backbone."
Expand Down
Loading