modelscope / modelscope/DiffSynth-Studio

Do the trained LoRA weights for Qwen-Image differ from those used in Diffusers? Why can't they be loaded properly in Diffusers?

Open
#886 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.1k
Forks
1.3k
Avg merge
13h 12m
Merged PRs (30d)
45

Description

I added the prefix “transformer.” to the keys and load Qwen-Image lora weights,but there is warning
Loading adapter weights from state_dict led to unexpected keys found in the model: transformer_blocks.0.img_mod.1.lora_A.default_0.default.weight, transformer_blocks.0.img_mod.1.lora_B.default_0.default.weight, transformer_blocks.0.attn.to_q.lora_A.default_0.default.weight, transformer_blocks.0.attn.to_q.lora_B.default_0.default.weight, transformer_blocks.0.attn.to_k.lora_A.default_0.default.weight, transformer_blocks.0.attn.to_k.lora_B.default_0.default.weight, transformer_blocks.0.attn.to_v.lora_A.default_0.default.weight, transformer_blocks.0.attn.to_v.lora_B.default_0.default.weight, transformer_blocks.0.attn.add_k_proj.lora_A.default_0.default.weight, transformer_blocks.0.attn.add_k_proj.lora_B.default_0.default.weight, transformer_blocks.0.attn.add_v_proj.lora_A.default_0.default.weight, transformer_blocks.0.attn.add_v_proj.lora_B.default_0.default.weight, transformer_blocks.0.attn.add_q_proj.lora_A.default_0.default.weight,

how to load lora weights properly in Diffusers.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the Qwen-Image LoRA loading warning in Diffusers and compare the reported state_dict keys with the model's expected keys. Trace whether the trained weights use a different naming or adapter format; done means the weights load without unexpected-key warnings and the behavior is verified.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.