Skip to content

Fix Qwen2.5-VL huggingface conversion issue (#2107) #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

Merged
meatybobby merged 6 commits into from
Feb 4, 2026
Merged

Fix Qwen2.5-VL huggingface conversion issue (#2107) #2156

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
a999acf
Fix Qwen2.5-VL conversion
meatybobby Jan 30, 2026
6ae2d75
Use recursive check for module
meatybobby Feb 2, 2026
f0a3f8b
Merge branch 'main' into bobchen/fix_qwen25
meatybobby Feb 2, 2026
9d9cc6b
Fix test
meatybobby Feb 2, 2026
eab31f2
Fix test
meatybobby Feb 2, 2026
23c68ad
Add test
meatybobby Feb 3, 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
2 changes: 2 additions & 0 deletions src/megatron/bridge/inference/vlm/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,6 +131,8 @@ def setup_inference_wrapper(
wrapper_cls = QwenVLInferenceWrapper
if isinstance(config, Qwen25VLModelProvider):
hidden_size = config.hidden_size
# Expose decoder for MCore Infernce Engine compatibility (used by get_mamba_inference_state_config_from_model)
mcore_model.module.decoder = mcore_model.module.language_model.decoder
Comment thread
meatybobby marked this conversation as resolved.
Outdated
else:
hidden_size = config.language_transformer_config.hidden_size
else:
Expand Down
3 changes: 0 additions & 3 deletions src/megatron/bridge/models/qwen_vl/modeling_qwen25_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,9 +111,6 @@ def __init__(
self.share_embeddings_and_output_weights = config.share_embeddings_and_output_weights
self.shared_embedding_or_output_weight = self.language_model.shared_embedding_or_output_weight

# Expose decoder for MCore Infernce Engine compatibility (used by get_mamba_inference_state_config_from_model)
self.decoder = self.language_model.decoder

# Bind methods from HF's Qwen2_5_VLModel to this instance
# get_placeholder_mask is only available in transformers 4.55+
if is_transformers_min_version("4.55.0"):
Expand Down
Loading