新增映射文档 #7081

Xuxuanang · 2025-03-06T01:16:08Z

新增转换规则 PaConvert#555

paddle-bot · 2025-03-06T01:16:14Z

感谢你贡献飞桨文档，文档预览构建中，Docs-New 跑完后即可预览，预览链接：http://preview-pr-7081.paddle-docs-preview.paddlepaddle.org.cn/documentation/docs/zh/api/index_cn.html
预览工具的更多说明，请参考：飞桨文档预览工具

docs/guides/model_convert/convert_from_pytorch/api_difference/cuda/torch.cuda.device_of.md

docs/guides/model_convert/convert_from_pytorch/api_difference/cuda/torch.cuda.get_rng_state.md

...nvert/convert_from_pytorch/api_difference/distributed/torch.distributed.monitored_barrier.md

...nvert_from_pytorch/api_difference/nn/torch.nn.modules.module.register_module_forward_hook.md

docs/guides/model_convert/convert_from_pytorch/api_difference/torch/torch.get_default_device.md

zhwesky2010 · 2025-03-10T04:01:26Z

docs/guides/model_convert/convert_from_pytorch/api_difference/cuda/torch.cuda.get_rng_state.md

+x = torch.cuda.get_rng_state(device='cuda：0')
+
+# Paddle 写法，返回 GeneratorState 对象
+x = paddle.get_cuda_rng_state()[0]


如果device无输入的话，应该是取第一个值吗？paddle.get_cuda_rng_state()[0]

这个在Matcher部分又重新修改了一下

...nvert/convert_from_pytorch/api_difference/distributed/torch.distributed.monitored_barrier.md

zhwesky2010 · 2025-03-10T08:05:00Z

上面的改了吗

docs/guides/model_convert/convert_from_pytorch/api_difference/cuda/torch.cuda.device_of.md

zhwesky2010 · 2025-03-12T03:33:25Z

docs/guides/model_convert/convert_from_pytorch/api_difference/cuda/torch.cuda.get_rng_state.md

+
+```python
+# PyTorch 写法，返回 torch.ByteTensor
+x = torch.cuda.get_rng_state(device='cuda：0')


这个是对device的转写，再加一个对返回参数的转写。对返回参数的转写，不需要传入device参数。

zhwesky2010 · 2025-03-12T03:38:06Z

...nvert_from_pytorch/api_difference/nn/torch.nn.modules.module.register_module_forward_hook.md

+| hook  | hook       | 被注册为 forward pre-hook 的函数。 |
+| prepend  | -            | 钩子执行顺序控制，Paddle 无此参数，暂无转写方式。 |
+| with_kwargs        | -       | 是否传递关键字参数，Paddle 无此参数，暂无转写方式。  |
+| always_call        | -       | 是否强制调用钩子，Paddle 无此参数，暂无转写方式。  |


prepend会控制pre、post吗？always_call会导致不调用吗？with_kwargs影响参数传递方式吗？

得看看这三个参数的影响是什么，是否会导致paddle的运行结果不同，如果对运行结果无影响就不用管直接删除了

这三个参数应该都有可能会导致运行结果不同，prepend控制当前注册的 hook 相对于已有 hook 的调用顺序，always_call如果设置为True的话，则在抛出异常会调用，with_kwargs会影响传递给hook的参数

zhwesky2010 · 2025-03-12T03:39:18Z

docs/guides/model_convert/convert_from_pytorch/pytorch_api_mapping_cn.md

@@ -1040,16 +1041,8 @@
 | 序号 | Pytorch 最新 release | Paddle develop | 映射关系分类 | 备注 |
 | ----- | ----------- | ----------------- | ----------- | ------- |
 | IN-DEVELOPMENT-PATTERN(`torch.nn.parameter.UninitializedParameter`, https://pytorch.org/docs/stable/generated/torch.nn.parameter.UninitializedParameter.html#torch.nn.parameter.UninitializedParameter) |
-| IN-DEVELOPMENT-PATTERN(`torch.nn.modules.module.register_module_forward_pre_hook`, https://pytorch.org/docs/stable/generated/torch.nn.modules.module.register_module_forward_pre_hook.html#torch-nn-modules-module-register-module-forward-pre-hook) |
-| IN-DEVELOPMENT-PATTERN(`torch.nn.modules.module.register_module_forward_hook`, https://pytorch.org/docs/stable/generated/torch.nn.modules.module.register_module_forward_hook.html#torch-nn-modules-module-register-module-forward-hook) |
-| IN-DEVELOPMENT-PATTERN(`torch.cuda.device_of`, https://pytorch.org/docs/stable/generated/torch.cuda.device_of.html#torch.cuda.device_of) |


这个分类不对，另外放到功能缺失里，你得细分一下类。

功能缺失也有好几类。

zhwesky2010 · 2025-03-12T04:07:32Z

torch.cuda.device_of这个还需要修改映射文档，对齐PaConvert

Xuxuanang · 2025-03-12T04:12:20Z

torch.cuda.device_of这个还需要修改映射文档，对齐PaConvert

这个不是已经删除，对应到功能缺失了吗

zhwesky2010

LGTM

zhwesky2010 · 2025-03-13T04:09:47Z

...nvert_from_pytorch/api_difference/nn/torch.nn.modules.module.register_module_forward_hook.md

@@ -0,0 +1,22 @@
+## [ torch 参数更多 ]torch.nn.modules.module.register_module_forward_hook


有几处错误：

torch的这两个API是给全局所有layer注册的hook，而paddle是单个layer的，这个差异必须要强调
torch.nn.modules.module.register_module_forward_hook
torch.nn.modules.module.register_module_forward_pre_hook

我查阅文档，没有发现register_module_forward_hook的prepend参数，这个参数是怎么来的？
https://pytorch.org/docs/main/generated/torch.nn.modules.module.register_module_forward_hook.html#torch-nn-modules-module-register-module-forward-hook

always_call我觉得不影响计算的结果，可以算作直接删除

这么看这两个API是无法自动转换的，可以写映射文档但Matcher是无法实现的（因为paddle没有类似的全局注册的API，只有单个注册的，没法将全局自动转为单个的逐个注册）。而PaConvert写的也有问题，由于单测是错的，直接用的Module.register_forward_hook，压根就没跑torch.nn.modules.module.register_module_forward_hook，导致两个错误Matcher没有测试到。

后续还是要保证代码的正确性。这些API只是初筛认为有对应功能，如果没有对应功能，就按你思考与判断的结论来处理，而不是一定要弄个错误的上去。

Xuxuanang added 8 commits December 27, 2024 22:45

fix mapping

ed34bc2

Merge branch 'PaddlePaddle:develop' into fix-docs

1d77aac

fix-docs

a617301

fix-docs

b7f12ca

fix-docs

2be1f0a

fix-docs

467f091

Merge branch 'PaddlePaddle:develop' into fix-docs

4f73037

update docs

ecb30a7

paddle-bot bot added the contributor label Mar 6, 2025

Xuxuanang added 2 commits March 6, 2025 09:42

fix-bugs

2995869

update docs

d9240e6

zhwesky2010 reviewed Mar 7, 2025

View reviewed changes

update docs

e571b61

zhwesky2010 reviewed Mar 10, 2025

View reviewed changes

Xuxuanang mentioned this pull request Mar 10, 2025

新增转换规则 PaddlePaddle/PaConvert#555

Merged

update docs

9aae6cd

Xuxuanang added 2 commits March 10, 2025 16:35

update docs

e1590b8

update docs

fa80805

Xuxuanang closed this Mar 11, 2025

Xuxuanang reopened this Mar 11, 2025

zhwesky2010 reviewed Mar 12, 2025

View reviewed changes

update docs

f08ac92

zhwesky2010 approved these changes Mar 13, 2025

View reviewed changes

zhwesky2010 merged commit fb9f287 into PaddlePaddle:develop Mar 13, 2025
2 checks passed

zhwesky2010 reviewed Mar 13, 2025

View reviewed changes

Xuxuanang deleted the fix-docs branch March 17, 2025 08:46

zhwesky2010 changed the title ~~Update docs~~ 映射文档补充与修改 Mar 31, 2025

zhwesky2010 changed the title ~~映射文档补充与修改~~ 新增映射文档 Mar 31, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

新增映射文档 #7081

新增映射文档 #7081

Xuxuanang commented Mar 6, 2025

paddle-bot bot commented Mar 6, 2025

zhwesky2010 Mar 10, 2025

Xuxuanang Mar 10, 2025

zhwesky2010 commented Mar 10, 2025

zhwesky2010 Mar 12, 2025 •

edited

Loading

zhwesky2010 Mar 12, 2025

Xuxuanang Mar 12, 2025 •

edited

Loading

zhwesky2010 Mar 12, 2025

zhwesky2010 commented Mar 12, 2025 •

edited

Loading

Xuxuanang commented Mar 12, 2025

zhwesky2010 left a comment

zhwesky2010 Mar 13, 2025 •

edited

Loading

		@@ -0,0 +1,22 @@
		## [ torch 参数更多 ]torch.nn.modules.module.register_module_forward_hook

新增映射文档 #7081

新增映射文档 #7081

Conversation

Xuxuanang commented Mar 6, 2025

paddle-bot bot commented Mar 6, 2025

zhwesky2010 Mar 10, 2025

Choose a reason for hiding this comment

Xuxuanang Mar 10, 2025

Choose a reason for hiding this comment

zhwesky2010 commented Mar 10, 2025

zhwesky2010 Mar 12, 2025 • edited Loading

Choose a reason for hiding this comment

zhwesky2010 Mar 12, 2025

Choose a reason for hiding this comment

Xuxuanang Mar 12, 2025 • edited Loading

Choose a reason for hiding this comment

zhwesky2010 Mar 12, 2025

Choose a reason for hiding this comment

zhwesky2010 commented Mar 12, 2025 • edited Loading

Xuxuanang commented Mar 12, 2025

zhwesky2010 left a comment

Choose a reason for hiding this comment

zhwesky2010 Mar 13, 2025 • edited Loading

Choose a reason for hiding this comment

zhwesky2010 Mar 12, 2025 •

edited

Loading

Xuxuanang Mar 12, 2025 •

edited

Loading

zhwesky2010 commented Mar 12, 2025 •

edited

Loading

zhwesky2010 Mar 13, 2025 •

edited

Loading