转换规则 No.243-300 其他未领取项 #217

co63oc · 2023-08-02T11:12:33Z

PR Docs

PR APIs

243 torch.linalg.inv_ex 功能缺失
246 torch.nn.LazyBatchNorm1d 功能缺失
247 torch.Tensor.sparse_resize_and_clear_ 功能缺失
248 torch.linalg.svdvals
251 torch.linalg.matrix_norm 功能缺失
254 torch.gradient 功能缺失
255 torch.special.gammaln
257 torch.nn.utils.parametrize.is_parametrized 功能缺失
259 torch.special.erfc 组合替代实现
260 torch.linalg.vector_norm 功能缺失
262 torch.nn.functional.group_norm 功能缺失
265 torch.special.expit 组合替代实现
266 torch.nn.Mish 已有文档验证无误
268 torch.linalg.cholesky_ex 功能缺失
269 torch.Tensor.sparse_resize_ 功能缺失
270 torch.positive 功能缺失
275 torch.special.exp2 组合替代实现
276 torch.nn.utils.parametrizations.spectral_norm
277 torch.cuda.StreamContext 功能缺失
278 torch.nn.LazyBatchNorm2d 功能缺失
281 torch.autograd.profiler.profile.self_cpu_time_total 功能缺失
284 torch.profiler.ProfilerActivity 功能缺失
286 torch.special.entr 功能缺失
287 torch.nn.LazyBatchNorm3d 功能缺失
288 torch.autograd.function.FunctionCtx.mark_dirty 功能缺失
289 torch.autograd.function.FunctionCtx.mark_non_differentiable
290 torch.autograd.Function.forward
291 torch.autograd.function.Function
292 torch.autograd.function.FunctionCtx.set_materialize_grads
293 torch.autograd.function.FunctionCtx.save_for_backward
294 torch.autograd.Function.backward
295 torch.profiler.ProfilerAction 功能缺失
296 torch.cuda.set_stream
300 torch.autograd.graph.saved_tensors_hooks

paddle-bot · 2023-08-02T11:12:40Z

Thanks for your contribution!

co63oc · 2023-08-02T12:29:50Z

FunctionCtx对象方法不能识别，torch.autograd.function.FunctionCtx.mark_dirty 增加一行代码用来识别

tests/test_autograd_Function_backward.py

tests/test_autograd_function_FunctionCtx_mark_dirty.py

tests/test_autograd_function_FunctionCtx_mark_non_differentiable.py

zhwesky2010 · 2023-08-03T08:49:08Z

tests/test_autograd_function_FunctionCtx_set_materialize_grads.py

+                return x+x+x, x+x
+
+            @staticmethod
+            def backward(ctx, grad, grad2):


可以仿照 https://pytorch.org/docs/stable/generated/torch.autograd.function.FunctionCtx.set_materialize_grads.html?highlight=set_materialize_grads#torch.autograd.function.FunctionCtx.set_materialize_grads
把case写严格一点，这个好像不是set_materialize_grads的用法

saved_tensors 没有合适转换方式，ctx不能识别

不能识别应该也会原封不动，这里如果函数名一致是不是也不会出错

paddle里是saved_tensor()，不是属性，不转换调用会有错误

tests/test_cuda_StreamContext.py

tests/test_gradient.py

zhwesky2010 · 2023-08-03T09:05:33Z

@co63oc 另外还需要处理下冲突

co63oc · 2023-08-03T11:23:39Z

@co63oc 另外还需要处理下冲突

已修改

zhwesky2010

后面写新增Matcher时可以按这个标准来哈：

zhwesky2010 · 2023-08-03T13:50:20Z

paconvert/api_matcher.py

+        else:
+            API_TEMPLATE = textwrap.dedent(
+                """
+                _, {}, _ = {}


直接写{}[1] 可以吧，省一行

如果转换后代码末尾为[]符号，测试框架会不转换api ，所以这里增加一行取数组元素，然后赋值

在 paconvert/transformer/basic_transformer.py:337 这里加一下 ast.Subscript 就可以

zhwesky2010 · 2023-08-03T13:55:12Z

tests/test_autograd_function_FunctionCtx_set_materialize_grads.py

+                return x+x+x, x+x
+
+            @staticmethod
+            def backward(ctx, grad, grad2):


不能识别应该也会原封不动，这里如果函数名一致是不是也不会出错

zhwesky2010 · 2023-08-03T13:56:38Z

tests/test_linalg_svdvals.py

+        """
+        import torch
+        x = torch.tensor([[12., -51, 4], [6, 167, -68], [-4, 24, -41]])
+        result = torch.linalg.svdvals(x, driver=None)


driver是属于 暂无转写方式 还是 可直接删除？

driver是可直接删除，已修改文档

zhwesky2010 · 2023-08-03T13:57:18Z

tests/test_nn_utils_parametrizations_spectral_norm.py

+        check_dtype=True,
+        check_stop_gradient=True,
+    ):
+        assert isinstance(paddle_result, paddle.nn.Linear)


这个映射是计算结果对不上吗？那就可能映射的不对了

返回类型不是Tensor类型，所以用isinstance判断

co63oc · 2023-08-04T14:09:23Z

后面写新增Matcher时可以按这个标准来哈：

好的

zhwesky2010 · 2023-08-15T09:16:05Z

paconvert/api_matcher.py

+        else:
+            API_TEMPLATE = textwrap.dedent(
+                """
+                _, {}, _ = {}


这个地方改成 {}[1]，在 paconvert/transformer/basic_transformer.py:337 这里加一下 ast.Subscript 就可以通过了，可以省去一行

zhwesky2010

LGTM

paddle-bot bot added contributor External developers status: proposed labels Aug 2, 2023

co63oc mentioned this pull request Aug 2, 2023

映射文档 No.243-300 其他未领取项 PaddlePaddle/docs#6075

Merged

co63oc force-pushed the api255 branch from 772aef0 to f2e9eef Compare August 2, 2023 11:38

luotao1 added the HappyOpenSource 快乐开源活动issue与PR label Aug 3, 2023

luotao1 assigned luotao1 and zhwesky2010 Aug 3, 2023

luotao1 mentioned this pull request Aug 3, 2023

Pytorch-Paddle代码转换工具开源任务 #112

Closed

zhwesky2010 reviewed Aug 3, 2023

View reviewed changes

co63oc force-pushed the api255 branch from 9c61e53 to f50a0ff Compare August 3, 2023 11:23

zhwesky2010 reviewed Aug 3, 2023

View reviewed changes

co63oc force-pushed the api255 branch from f50a0ff to 2ce39b7 Compare August 4, 2023 14:03

Add tests

d89ae4b

co63oc force-pushed the api255 branch from 2ce39b7 to d89ae4b Compare August 4, 2023 14:08

co63oc added 3 commits August 4, 2023 22:30

Fix

ecc61cf

Merge branch 'master' into api255

0e31b09

Fix

4aaa69a

zhwesky2010 reviewed Aug 15, 2023

View reviewed changes

zhwesky2010 approved these changes Aug 15, 2023

View reviewed changes

zhwesky2010 merged commit 6c56024 into PaddlePaddle:master Aug 15, 2023
8 checks passed

co63oc deleted the api255 branch August 19, 2023 00:07

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

转换规则 No.243-300 其他未领取项 #217

转换规则 No.243-300 其他未领取项 #217

co63oc commented Aug 2, 2023 •

edited

paddle-bot bot commented Aug 2, 2023

co63oc commented Aug 2, 2023

zhwesky2010 Aug 3, 2023

co63oc Aug 3, 2023

zhwesky2010 Aug 3, 2023

co63oc Aug 4, 2023 •

edited

zhwesky2010 commented Aug 3, 2023

co63oc commented Aug 3, 2023

zhwesky2010 left a comment

zhwesky2010 Aug 3, 2023

co63oc Aug 4, 2023

zhwesky2010 Aug 14, 2023

zhwesky2010 Aug 3, 2023

zhwesky2010 Aug 3, 2023

co63oc Aug 4, 2023

zhwesky2010 Aug 3, 2023

co63oc Aug 4, 2023

co63oc commented Aug 4, 2023

zhwesky2010 Aug 15, 2023

zhwesky2010 left a comment

转换规则 No.243-300 其他未领取项 #217

转换规则 No.243-300 其他未领取项 #217

Conversation

co63oc commented Aug 2, 2023 • edited

PR Docs

PR APIs

paddle-bot bot commented Aug 2, 2023

co63oc commented Aug 2, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

co63oc Aug 4, 2023 • edited

Choose a reason for hiding this comment

zhwesky2010 commented Aug 3, 2023

co63oc commented Aug 3, 2023

zhwesky2010 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

co63oc commented Aug 4, 2023

Choose a reason for hiding this comment

zhwesky2010 left a comment

Choose a reason for hiding this comment

co63oc commented Aug 2, 2023 •

edited

co63oc Aug 4, 2023 •

edited