【Hackathon 7th No.37】为 Paddle 代码转换工具新增 API 转换规则（第 4 组） #481

monster1015 · 2024-09-21T02:39:56Z

PR Docs

PR APIs

torch.mvlgamma
torch.Tensor.igamma
torch.Tensor.igamma_
torch.Tensor.igammac
torch.Tensor.igammac_
torch.Tensor.mvlgamma
torch.Tensor.mvlgamma_
torch.ormqr
torch.Tensor.ormqr
torch.Tensor.orgqr
torch.special.ndtr
torch.Tensor.matrix_exp
torch.linalg.inv_ex
torch.linalg.cholesky_ex
torch._assert
torch.testing.make_tensor

monster1015 · 2024-09-21T05:52:12Z

请问如何查看CI的具体结果？我点进去是一片白的 @luotao1

luotao1 · 2024-09-23T02:06:33Z

请问如何查看CI的具体结果？我点进去是一片白的

cc @tianshuo78520a @risemeup1

luotao1 · 2024-09-23T03:33:38Z

@monster1015 日志已打开

monster1015 · 2024-09-23T08:06:24Z

结果不是对得上吗？还有就是torch.Tensor.ormqr就是有至少两个输入的 @luotao1 @zhwesky2010

monster1015 · 2024-09-23T08:44:03Z

还有这个，self.paddleClass不是获取对应输入的？

zhwesky2010 · 2024-10-08T11:03:37Z

paconvert/api_mapping.json

+      "out"
+    ],
+    "unsupport_args": [
+      "check_errors"


文档改了这里不改？

zhwesky2010 · 2024-10-08T11:03:43Z

paconvert/api_mapping.json

+      "out"
+    ],
+    "unsupport_args": [
+      "check_errors"


文档改了这里不改？

zhwesky2010 · 2024-10-08T11:05:15Z

paconvert/api_mapping.json

+    "unsupport_args": [
+      "noncontiguous",
+      "exclude_zero",
+      "memory_format"


这些其实没必要全部算作无法支持，只有影响到网络计算的才算。

zhwesky2010 · 2024-10-08T11:08:20Z

paconvert/api_matcher.py

+        return code
+
+
+class TensorMatrixExpMatcher(BaseMatcher):


这个matcher能否复用class TensorFunc2PaddleFunc

zhwesky2010 · 2024-10-08T11:10:39Z

paconvert/api_matcher.py

+        else:
+            kwargs["x"] = kwargs["input"]
+        if "left" not in kwargs.keys():
+            kwargs["left"] = True


这些参数默认不是True吗，还是说默认参数不同，必须要设置一下

建议转写代码尽可能简洁无冗余

参数是一致的，我已经修改了，docs和对应的mathcer也已经修改

zhwesky2010

丰富下测试case，以下四种情况的测试case必须全部包含：

传入所有参数且全部指定关键字
传入所有参数且全部不指定关键字
改变关键字顺序
默认参数均不指定

zhwesky2010 · 2024-10-09T08:32:10Z

paconvert/api_mapping.json

+    "min_input_args": 1,
+    "args_list": [
+      "A",
+      "check_errors",


这个不对吧，放在*的前后意义是不同的，是 位置参数 还是 指定关键字参数？

他是指定关键字参数

zhwesky2010 · 2024-10-09T08:33:45Z

paconvert/api_mapping.json

-  "torch.Tensor.orgqr": {},
-  "torch.Tensor.ormqr": {},
+  "torch.Tensor.orgqr": {
+    "Matcher": "OrgqrMatcher",


命名TensorOrgqrMatcher吧

zhwesky2010 · 2024-10-09T08:39:09Z

tests/test_linalg_cholesky_ex.py

+        [0.74611458, 1.24800785, 0.88039371]])
+        out1 = torch.randn(3, 3)
+        info1 = torch.tensor([1, 2, 3], dtype=torch.int32)
+        torch.linalg.cholesky_ex(a, check_errors=False, upper=True, out=(out1, info1))


check_errors=True测一下

zhwesky2010 · 2024-10-09T08:39:21Z

tests/test_linalg_inv_ex.py

+                [0.48016702, 0.14235102, 0.42620817]])
+        out1 = torch.tensor([])
+        info1 = torch.tensor([1, 2, 3], dtype=torch.int32)
+        out1, info1 = torch.linalg.inv_ex(x, check_errors=False, out=(out1, info1))


check_errors=True的情况也测下

已经完善测试case了

monster1015 · 2024-10-10T05:54:15Z

已经全都跑通了 @zhwesky2010

zhwesky2010 · 2024-10-11T08:03:33Z

paconvert/api_matcher.py

+        API_TEMPLATE = textwrap.dedent(
+            """
+            out = paddle.uniform({}, dtype={}, min={}, max={}).to({})
+            out.stop_gradient = not {}


如果没有requires_grad，这个地方可以简写为一行
有requires_grad才需要扩充为三行

参考下GenericMatcher

zhwesky2010 · 2024-10-11T08:08:34Z

paconvert/api_matcher.py

+class LinalgCholeskyExMatcher(BaseMatcher):
+    def generate_code(self, kwargs):
+        if "upper" not in kwargs:
+            kwargs["upper"] = False


这个一定要单独设置吗？默认值不就是False吗，直接self.kwargs_to_str不就行了吗

主要是为了配合下面的代码

zhwesky2010 · 2024-10-11T08:15:22Z

paconvert/api_mapping.json

+    ]
+  },
+  "torch.Tensor.ormqr": {
+    "Matcher": "OrmqrMatcher",


这个配置TensorFunc2PaddleFunc就可以吧

这个可以

zhwesky2010 · 2024-10-11T08:15:32Z

paconvert/api_mapping.json

@@ -13933,6 +14051,18 @@
      "verbose"
    ]
  },
+  "torch.ormqr": {
+    "Matcher": "OrmqrMatcher",


这个配置GenericMatcher就可以吧

这个也可以

zhwesky2010 · 2024-10-11T08:16:07Z

paconvert/api_matcher.py

+            return code
+
+
+class OrmqrMatcher(BaseMatcher):


这个逻辑简单应该不用单独写Matcher了，分别配置 TensorFunc2PaddleFunc和GenericMatcher 就行

这个我觉得应该设置的，其他是已经改完了的。请review

zhwesky2010 · 2024-10-12T06:44:05Z

paconvert/api_matcher.py

@@ -807,6 +807,79 @@ def generate_code(self, kwargs):
        return code


+class TensorOrgqrMatcher(BaseMatcher):


这个也可以用GenericMatcher，配置一个kwargs_change

zhwesky2010 · 2024-10-12T06:46:56Z

paconvert/api_mapping.json

+    "Matcher": "MakeTMatcher",
+    "min_input_args": 3,
+    "args_list": [
+      "shape",


这个看起来shape是 可变参数 或 位置参数

其他的应该是 指定关键字参数

所以应该这么写：

"args_list": [ "*shape", "*", "device", "dtype", "low", "high", "requires_grad", "noncontiguous", "exclude_zero", "memory_format" ]

只看torch文档可能不精确

zhwesky2010 · 2024-10-12T06:50:35Z

paconvert/api_matcher.py

+                """
+            )
+            code = API_TEMPLATE.format(
+                kwargs["shape"],


这里面的这个shape比较复杂，因为其既可以是 可变参数：

torch.testing.make_tensor(3, 2, dtype=torch.float32, device=torch.device('cpu'))

也可以是 位置参数：

torch.testing.make_tensor([1, 2], dtype=torch.float32, device=torch.device('cpu'))

参考下CreateMatcher的写法吧，如果比较复杂不会写，可以看看是不是复用CreateMatcher，再配置下json字段就行

好的好的，我试试

monster1015 · 2024-10-12T09:05:46Z

已经好了，请review @zhwesky2010

zhwesky2010

LGTM

【Hackathon 7th No.37】为 Paddle 代码转换工具新增 API 转换规则（第 4 组）

3ba9d2c

paddle-bot bot added the contributor External developers label Sep 21, 2024

luotao1 mentioned this pull request Sep 23, 2024

【Hackathon 7th】开源贡献个人挑战赛 PaddlePaddle/Paddle#68244

Open

luotao1 assigned luotao1 and zhwesky2010 Sep 23, 2024

update

754006b

monster1015 added 2 commits September 23, 2024 16:10

update

376663d

update

7c49e1f

monster1015 and others added 8 commits September 23, 2024 20:43

update

699cb20

update

4ac7e22

Merge branch 'PaddlePaddle:master' into master

74ad029

update

013c7fb

update

1a0fe37

update

e955a5a

update

dca6539

update

bd74a9c

zhwesky2010 reviewed Oct 8, 2024

View reviewed changes

monster1015 and others added 4 commits October 9, 2024 13:01

update

e0362e7

update

71f7f77

update

8d7fe8f

Merge branch 'master' into master

d438a29

zhwesky2010 reviewed Oct 9, 2024

View reviewed changes

monster1015 added 3 commits October 10, 2024 10:34

update

0c4c3d8

Merge branch 'master' of LFM:monster1015/PaConvert

9faff94

update

23d1aa3

monster1015 added 2 commits October 10, 2024 10:54

update

12d9e64

update

9596eb7

zhwesky2010 reviewed Oct 11, 2024

View reviewed changes

update

51743b8

zhwesky2010 reviewed Oct 12, 2024

View reviewed changes

update

adea30b

zhwesky2010 approved these changes Oct 12, 2024

View reviewed changes

zhwesky2010 merged commit b6c624e into PaddlePaddle:master Oct 12, 2024
7 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

【Hackathon 7th No.37】为 Paddle 代码转换工具新增 API 转换规则（第 4 组） #481

【Hackathon 7th No.37】为 Paddle 代码转换工具新增 API 转换规则（第 4 组） #481

monster1015 commented Sep 21, 2024 •

edited

Loading

monster1015 commented Sep 21, 2024

luotao1 commented Sep 23, 2024

luotao1 commented Sep 23, 2024

monster1015 commented Sep 23, 2024

monster1015 commented Sep 23, 2024

zhwesky2010 Oct 8, 2024

zhwesky2010 Oct 8, 2024

zhwesky2010 Oct 8, 2024

zhwesky2010 Oct 8, 2024

zhwesky2010 Oct 8, 2024

monster1015 Oct 9, 2024

zhwesky2010 left a comment

zhwesky2010 Oct 9, 2024

monster1015 Oct 10, 2024

zhwesky2010 Oct 9, 2024

zhwesky2010 Oct 9, 2024

zhwesky2010 Oct 9, 2024

monster1015 Oct 10, 2024

monster1015 commented Oct 10, 2024

zhwesky2010 Oct 11, 2024

zhwesky2010 Oct 11, 2024

monster1015 Oct 11, 2024

zhwesky2010 Oct 11, 2024

monster1015 Oct 11, 2024

zhwesky2010 Oct 11, 2024

monster1015 Oct 11, 2024

zhwesky2010 Oct 11, 2024

monster1015 Oct 11, 2024

zhwesky2010 Oct 12, 2024

zhwesky2010 Oct 12, 2024

zhwesky2010 Oct 12, 2024

monster1015 Oct 12, 2024

monster1015 commented Oct 12, 2024

zhwesky2010 left a comment

		@@ -807,6 +807,79 @@ def generate_code(self, kwargs):
		return code


		class TensorOrgqrMatcher(BaseMatcher):

【Hackathon 7th No.37】为 Paddle 代码转换工具新增 API 转换规则（第 4 组） #481

【Hackathon 7th No.37】为 Paddle 代码转换工具新增 API 转换规则（第 4 组） #481

Conversation

monster1015 commented Sep 21, 2024 • edited Loading

PR Docs

PR APIs

monster1015 commented Sep 21, 2024

luotao1 commented Sep 23, 2024

luotao1 commented Sep 23, 2024

monster1015 commented Sep 23, 2024

monster1015 commented Sep 23, 2024

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

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

monster1015 commented Oct 10, 2024

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

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

monster1015 commented Oct 12, 2024

zhwesky2010 left a comment

Choose a reason for hiding this comment

monster1015 commented Sep 21, 2024 •

edited

Loading