-
Notifications
You must be signed in to change notification settings - Fork 73
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[WeeklyReport] Tsaiyue 2024.04.06~2024.04.19 (#239)
Co-authored-by: westfish <westfish@126.com>
- Loading branch information
Showing
1 changed file
with
36 additions
and
0 deletions.
There are no files selected for viewing
36 changes: 36 additions & 0 deletions
36
WeeklyReports/Hackathon_6th/22_Tsaiyue/[WeeklyReports]2024.04.06~2024.04.19.md
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,36 @@ | ||
### 姓名 | ||
|
||
蔡越 | ||
|
||
### 实习项目 | ||
|
||
PaddleMIX 套件能力建设(文图方向) | ||
|
||
### 本周工作 | ||
|
||
1. **优化AnimateAnyone训练权重保存逻辑以即混合精度控制机制;** | ||
|
||
- 根据两阶段权重保存逻辑,重写trainer中_save()函数,实现在training_args参数控制下对模型相关信息进行保存; | ||
|
||
- 清除原始精度控制机制,统一改由trainer进行精度策略控制,并在不同精度混合进度策略下进行测试; | ||
|
||
- 重构模型相关代码,将其融合进ppdiffusers/models下; | ||
|
||
- 关于模型组网denoising_unet()的权重加载方式,由于其包含对motion_module的更新,与ppdiffusers的模型权重统一加载方式难以兼容,考虑先保留现有的权重加载方式; | ||
|
||
- 相关PR:https://github.com/PaddlePaddle/PaddleMIX/pull/501 , https://github.com/PaddlePaddle/PaddleMIX/pull/510 | ||
|
||
2. **同步更新AI studio项目** | ||
|
||
- 解决由paddle2.5.2升级为2.6.1导致部分算子(transpose)计算行为发生变化的问题,目前已支持paddlepaddle2.6.1; | ||
|
||
- 项目链接:https://aistudio.baidu.com/projectdetail/7490749?contributionType=1 | ||
|
||
### 下周工作 | ||
|
||
1. 推进AnimateAnyone训练支持相关PR合入; | ||
2. 调研结合transformer和diffusion model的视频生成开源模型,与导师沟通规划下一步工作重点; | ||
|
||
### 导师点评 | ||
|
||
蔡越在本周围完成了AnimateAnyone的结尾工作,完成了相关PR的合入和ai studio项目的更新,并开展了open-sora相关的调研工作,干的不错,继续加油。 |