modelscope / modelscope/DiffSynth-Studio
关于Wan训练时的timestep
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 13.1k
- Forks
- 1.3k
- Avg merge
- 13h 12m
- Merged PRs (30d)
- 45
Description
https://github.com/modelscope/DiffSynth-Studio/blob/main/diffsynth/pipelines/wan_video_new.py#L108
这个函数中timestep_id的范围是[0, num_train_timesteps）也就是[0, 1000)，而self.scheduler.timesteps的长度只有num_inference_steps，默认值是100，而在实际推理的时候可能只有40/50。
也就是说，会生成一个0到1000的随机数作为一个长度为100的数组的索引，这不会出现问题吗？
期待您的解答，谢谢！
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
Start at diffsynth/pipelines/wan_video_new.py around line 108 and trace how timestep_id is generated and used with self.scheduler.timesteps. Compare the stated ranges and inference-step lengths, then verify the behavior in the Wan training and inference paths. Done means confirming whether the indexing is valid and documenting the explanation or identifying the precise change required.
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