-
Notifications
You must be signed in to change notification settings - Fork 72
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[WeeklyReport] NKNaN 2024.02.25~2024.03.08 #151
Conversation
|
||
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。 |
There was a problem hiding this comment.
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
There was a problem hiding this comment.
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 |
There was a problem hiding this comment.
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?
There was a problem hiding this 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)
There was a problem hiding this 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。 |
There was a problem hiding this comment.
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?
There was a problem hiding this comment.
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 |
There was a problem hiding this comment.
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。
@NKNaN 补充一下点评内容,然后我approve |
No description provided.