[Typing][B-40] Add type annotations for python/paddle/autograd/backward_mode.py
#66277
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
| def backward( | ||
| tensors: list[paddle.Tensor], | ||
| grad_tensors: list[paddle.Tensor] | None = None, | ||
| retain_graph: bool | None = False, | ||
| ) -> None: |
There was a problem hiding this comment.
几个小问题：
- 最好用
Tensor，TYPE_CHECKING中导入 grad_tensors的list元素可以有Noneretain_graph为啥有None？
There was a problem hiding this comment.
好的好的，我之前理解错optional的含义了，已更正
| @@ -32,7 +38,7 @@ def backward(tensors, grad_tensors=None, retain_graph=False): | |||
| If None, all the gradients of the ``tensors`` is the default value which is filled with 1.0. | |||
| Defaults to None. | |||
|
|
|||
| retain_graph(bool, optional): If False, the graph used to compute grads will be freed. If you would | |||
| retain_graph(bool|False, optional): If False, the graph used to compute grads will be freed. If you would | |||
There was a problem hiding this comment.
| retain_graph(bool|False, optional): If False, the graph used to compute grads will be freed. If you would | |
| retain_graph(bool, optional): If False, the graph used to compute grads will be freed. If you would |
不需要写具体值
| def check_tensors( | ||
| in_out_list: list[paddle.Tensor], name: str | ||
| ) -> list[paddle.Tensor]: |
There was a problem hiding this comment.
内部函数一般不用标注，如果要标注的话，这里其实有点问题，in_out_list 应该是 list[Tensor] | tuple[Tensor, ...] | Tensor ，输出也可以是 tuple
|
|
|
注意一下 CLA 的签署，git commit 所使用的 email 要和 GitHub 上的一致，否则 PR 是没办法合入的，你的所有 PR 都有这个问题 即便新的 commit 使用的 email 正确，过去的 commit 也会阻塞 CLA 签署，所以需要将过去的 commit 重新编辑（比如 rebase）后 force push |
| import paddle | ||
| from paddle.base import core, framework | ||
| from paddle.base.backward import gradients_with_optimizer # noqa: F401 | ||
|
|
||
| if TYPE_CHECKING: | ||
| from .. import Tensor |
There was a problem hiding this comment.
| from .. import Tensor | |
| from paddle import Tensor |
| def backward(tensors, grad_tensors=None, retain_graph=False): | ||
| def backward( | ||
| tensors: list[Tensor], | ||
| grad_tensors: list[Tensor, None] | None = None, |
There was a problem hiding this comment.
| grad_tensors: list[Tensor, None] | None = None, | |
| grad_tensors: list[Tensor | None] | None = None, |
list 只有一个泛型参数
| def check_tensors( | ||
| in_out_list: list[Tensor] | tuple[Tensor, ...] | Tensor, name: str | ||
| ) -> list[Tensor] | tuple[Tensor, ...]: |
There was a problem hiding this comment.
| def check_tensors( | |
| in_out_list: list[Tensor] | tuple[Tensor, ...] | Tensor, name: str | |
| ) -> list[Tensor] | tuple[Tensor, ...]: | |
| def check_tensors( | |
| in_out_list: Sequence[Tensor] | Tensor, name: str | |
| ) -> Sequence[Tensor]: |
另外，104、109、126 行的 isinstance(x, (paddle.Tensor, core.eager.Tensor)) 属于历史遗留写法，现在 paddle.Tensor 和 core.eager.Tensor 是同一个东西，可以合并为 isinstance(x, paddle.Tensor)
There was a problem hiding this comment.
上述内容已修正，并对代码部分做了更新，谢谢您的suggestion！
SigureMo
changed the title
python/paddle/autograd/backward_mode.py
PR Category
User Experience
PR Types
Improvements
Description
类型标注：
Related links
@SigureMo @megemini