Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

[WeeklyReport] NKNaN 2024.02.25~2024.03.08 #151

Merged
merged 3 commits into from
Mar 14, 2024
Merged

[WeeklyReport] NKNaN 2024.02.25~2024.03.08 #151

merged 3 commits into from
Mar 14, 2024

Conversation

NKNaN
Copy link
Contributor

@NKNaN NKNaN commented Mar 8, 2024

No description provided.

@NKNaN NKNaN assigned NKNaN and zhwesky2010 and unassigned NKNaN Mar 8, 2024
@NKNaN NKNaN added the weekly reports weekly report commits from members label Mar 8, 2024

2. `paddle.nanmedian`: 若输入类型为 int 类型,输出在取平均时是按照整数除法,导致输出不是精确的平均值。

3. `paddle.nn.functional.interpolate`: 在 `mode = nearest` 时不支持 3-D tensor 而 torch 是支持的,其他 `mode` 下可能也有兼容的 tensor 维度和 torch 不一致的情况,需要进一步考察。

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

其他问题多吗,这些是属于历史的 功能缺失 层面吗,如果是这样就暂不需要支持了

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

是属于历史的功能缺失层面,之前在群里提到的。


#### 备注——其他发现的问题

1. `paddle.nn.CrossEntropyLoss`: https://github.com/PaddlePaddle/Paddle/issues/62009

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个是什么问题?是框架问题还是看代码疑问?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

属于看代码的疑问吧,我觉得在 soft_label=True 且输入 weight 的情况下 paddle 和 torch 的计算不一致PaddlePaddle/Paddle#62009 (comment)

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

属于看代码的疑问吧,我觉得在 soft_label=True 且输入 weight 的情况下 paddle 和 torch 的计算不一致PaddlePaddle/Paddle#62009 (comment)

确定一下是否是框架Bug,如果是Bug的话就需要登记到表里作为后续工作计划


1. `paddle.nn.CrossEntropyLoss`: https://github.com/PaddlePaddle/Paddle/issues/62009

2. `paddle.nanmedian`: 若输入类型为 int 类型,输出在取平均时是按照整数除法,导致输出不是精确的平均值。

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

mode='avg'逻辑不变,不做不兼容升级,mode='min'时与torch一致,这样开发有问题吗?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

没有问题,目前是这样修改的。

1. 完善已提交的 pr
2. 继续修改其他框架 API

### 导师点评
Copy link

@zhwesky2010 zhwesky2010 Mar 12, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

导师点评:

李睿文修改了upsampling、ctc_loss、median、nonzero的API升级,但ctc_loss、nonzero的推进难度较大。在工作中需要注意,结合当前的问题点,分析其对应的工作量、风险等,要把握关键问题点,忽略其他无关问题,保持目标专注不要被无关细节困扰。同时工作前提前评估成本,先易后难的方式开展。对于不兼容升级,应考虑到其成本的倍增。

@zhwesky2010
Copy link

@NKNaN 补充一下点评内容,然后我approve

@AndSonder AndSonder merged commit 3debbec into PFCCLab:main Mar 14, 2024
1 check passed
# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
weekly reports weekly report commits from members
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants