-
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] NKNaN 2024.08.26~2024.09.08 (#374)
- Loading branch information
Showing
1 changed file
with
66 additions
and
0 deletions.
There are no files selected for viewing
66 changes: 66 additions & 0 deletions
66
WeeklyReports/Hackathon_7th/05_NKNaN/[WeeklyReports]2024.08.26~2024.09.08.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 +1,66 @@ | ||
### 姓名 | ||
|
||
李睿文 | ||
|
||
### 实习项目 | ||
|
||
框架 API 易用性提升 | ||
|
||
### 本周工作 | ||
|
||
1. **完成核对已修改的 58 个 API** | ||
|
||
- 核对已修改的 58 个 API 的 API 映射文档、matcher、单测。 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/PaConvert/pull/463 、https://github.com/PaddlePaddle/docs/pull/6848 、https://github.com/PaddlePaddle/Paddle/pull/67772 | ||
|
||
2. **修复 `paddle.diff` 存在的 bug** | ||
|
||
- 当 paddle.diff 输入 n!=1 且有 prepend 或 append 参数时返回结果与 pytorch 不一致。 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/Paddle/pull/67800 | ||
|
||
3. **增强 `paddle.optimizer.SGD/Adadelta/Adagrad/RMSprop` 等优化器的功能** | ||
|
||
- paddle.optimizer.SGD/Adadelta/Adagrad/RMSprop 等优化器的 weight_decay 参数需增加 int 输入类型的支持 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/Paddle/pull/67947、https://github.com/PaddlePaddle/Paddle/pull/68033 | ||
|
||
|
||
4. **修复 `paddle.linalg.matrix_rank` 新增 kernel 存在的 bug** | ||
|
||
- 修复当 rtol 输入为 None 的计算分支内关于 rtol_tensor 形状的问题 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/Paddle/pull/68108 | ||
|
||
5. **增强 `F.max_unpool1d/F.max_unpool2d/F.max_unpool3d` 输入 indices 的类型** | ||
|
||
- F.max_unpool1d/F.max_unpool2d/F.max_unpool3d 的输入 indices 需增加 int64 类型输入的支持 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/Paddle/pull/68046 | ||
|
||
6. **修复 `F.batch_norm` 存在的 bug** | ||
|
||
- 当 training=True 时,F.batch_norm 的 running_variance 计算结果与 pytorch 不同,因为 pytorch 使用的是无偏方差,paddle 是有偏方差 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/Paddle/pull/68159 | ||
|
||
7. **增强 `F.pad` 的 data_format 参数** | ||
|
||
- F.pad 的 data_format 需自动适配3D/5D输入 | ||
|
||
- pr 链接:https://github.com/PaddlePaddle/Paddle/pull/68214 | ||
|
||
#### 问题疑惑与解答 | ||
|
||
暂无 | ||
|
||
### 下周工作 | ||
|
||
1. 完善未合入的 pr 以及对应的映射文档、matcher、json 映射法则以及 PaConvert 中的单测; | ||
1. 分析 paddle.nonzero 修改的不兼容升级问题; | ||
1. paddle.load 增强输入类型; | ||
1. paddle.chunk 支持非整除情况 | ||
|
||
### 导师点评 | ||
李睿文同学高质量的完成了多个API的修正工作,同时快速掌握PaConvert的各种开发流程,针对历史修改的全部存量API完成了详细的核对。工作中细节严谨,后续继续注重细节,把握各项文档/单测/API/Matcher等开发规范。 |