modelscope / modelscope/DiffSynth-Studio
qwen-image distill模型训练的一些问题
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 13.1k
- Forks
- 1.3k
- Avg merge
- 13h 12m
- Merged PRs (30d)
- 45
Description
您好，我在阅读源码时有一些疑问，想请教一下：
蒸馏模型在训练时是直接通过 SFT（supervised fine-tuning）把 CFG 蒸掉的吗？
是否没有额外的蒸馏约束？
比如类似 DMD: Distilled Model Diffusion 这样的蒸馏方法？
我注意到在代码中，不论是「蒸馏模型」还是「非蒸馏模型」的训练流程，数据预处理后都没有使用 inputs_nega。
那么对于 非蒸馏模型 的训练，是否不应该有一定概率 drop prompt，从而通过 inputs_nega 来生成图像？
目前看起来 inputs_nega 并没有被利用。
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Inspect the training data-preprocessing and training paths for distilled and non-distilled Qwen-Image models, focusing on whether inputs_nega is referenced. Compare the observed behavior with the issue's SFT, CFG-drop, and DMD questions; completion requires establishing the intended training design and documenting or implementing an agreed correction.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100