【SCU】【Paddle Tensor No.25】新增paddle.vecdot , paddle.linalg.vecdot
#69477
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
python/paddle/tensor/linalg.py
Outdated
| [32, 64]) | ||
|
|
||
| """ | ||
| return dot(x, y) |
There was a problem hiding this comment.
参考API规范定义，vecdot应该是逐元素乘法后再在axis轴上求和：https://data-apis.org/array-api/latest/API_specification/generated/array_api.vecdot.html#array_api.vecdot
所以可以参考如下实现（复用乘法和sum）
https://github.com/pytorch/pytorch/blob/main/torch/_refs/linalg/__init__.py#L307
There was a problem hiding this comment.
omg sry，已修改，谢谢佬更正！
test/legacy_test/test_dot_op.py
Outdated
There was a problem hiding this comment.
-
这是新增的API，不是OP，参考这几个文件写一下单测
-
文件名建议改成test_linalg_vecdot.py
Co-authored-by: HydrogenSulfate <490868991@qq.com>
Co-authored-by: HydrogenSulfate <490868991@qq.com>
|
麻烦解决一下conflict |
| def test_dtype_mismatch(self): | ||
| with self.assertRaises(TypeError): | ||
| x = paddle.rand([3, 4], dtype="float32") | ||
| y = paddle.rand([3, 4], dtype="int32") | ||
| paddle.vecdot(x, y, axis=-1) |
There was a problem hiding this comment.
这个单测需要跳过XPU：
@unittest.skipIf(
core.is_compiled_with_xpu(),
"Skip XPU for not support uniform(dtype=int)",
)| paddle.vecdot(x, y, axis=-1) | ||
|
|
||
|
|
||
| class VecDotTestCaseComplex(unittest.TestCase): |
There was a problem hiding this comment.
这个类跳过XPU：
@unittest.skipIf(
core.is_compiled_with_xpu(),
"Skip XPU for not support complex",
)| ) | ||
|
|
||
|
|
||
| class VecDotTestCaseTypePromotion2(unittest.TestCase): |
There was a problem hiding this comment.
同上，跳过XPU，
@unittest.skipIf(
core.is_compiled_with_xpu(),
"Skip XPU for not support complex",
)
|
建议不要在网页上编辑，还是本地经过pre-commit比较好呢， |
python/paddle/tensor/linalg.py
Outdated
| >>> import paddle | ||
| >>> x = paddle.to_tensor([1, 2, 3], dtype='float32') | ||
| >>> y = paddle.to_tensor([4, 5, 6], dtype='float32') | ||
| >>> result = paddle.linalg.vecdot(x, y) | ||
| >>> print(result) | ||
| Tensor(shape=[], dtype=float32, place=Place(cpu), stop_gradient=True, | ||
| [32.0]) | ||
|
|
||
| >>> x2 = paddle.to_tensor([[1, 2, 3], [4, 5, 6]], dtype='float32') | ||
| >>> y2 = paddle.to_tensor([[1, 2, 3], [4, 5, 6]], dtype='float32') | ||
| >>> result2 = paddle.linalg.vecdot(x2, y2, axis=1) | ||
| >>> print(result2) | ||
| Tensor(shape=[2], dtype=float32, place=Place(cpu), stop_gradient=True, | ||
| [14.0, 77.0]) |
There was a problem hiding this comment.
看下样例代码的输出的空格缩进是否与实际输出完全一致：
|
@HydrogenSulfate 应该可以了，老师您再看看ww |
|
|
||
|
|
||
| class VecDotTestCaseTypePromotion2(unittest.TestCase): | ||
| def test_float64_complex64_promotion(self): |
There was a problem hiding this comment.
这个也是，所有complex类型的都加上对xpu的skip，否则CI过不了
python/paddle/tensor/linalg.py
Outdated
| >>> import paddle | ||
| >>> x = paddle.to_tensor([1, 2, 3], dtype='float32') | ||
| >>> y = paddle.to_tensor([4, 5, 6], dtype='float32') | ||
| >>> result = paddle.linalg.vecdot(x, y) | ||
| >>> print(result) | ||
| Tensor(shape=[], dtype=float32, place=Place(cpu), stop_gradient=True, | ||
| [32.0]) | ||
|
|
||
| >>> x2 = paddle.to_tensor([[1, 2, 3], [4, 5, 6]], dtype='float32') | ||
| >>> y2 = paddle.to_tensor([[1, 2, 3], [4, 5, 6]], dtype='float32') | ||
| >>> result2 = paddle.linalg.vecdot(x2, y2, axis=1) | ||
| >>> print(result2) | ||
| Tensor(shape=[2], dtype=float32, place=Place(cpu), stop_gradient=True, | ||
| [14.0, 77.0]) |
There was a problem hiding this comment.
样例代码的输出部分，缩进修改一下，不然example code检测挂了：
|
@PolaKuma 注意文档格式：尤其是缩进空格和末尾的小数点，建议手动运行一下样例代码，然后把终端里的输出粘贴上去 |
|
@PolaKuma 缩进空格也要相同，32的"3"要跟上面的"s"对齐 |
我换一个例子好了🥲我本地跑都没问题呜呜 |
HydrogenSulfate
changed the title
paddle.vecdot , paddle.linalg.vecdot，复用函数paddle.linalg.dotpaddle.vecdot , paddle.linalg.vecdot
PR Category
User Experience
PR Types
New features
Description
Paddle Tensor 规范化：新增
paddle.vecdot,paddle.linalg.vecdot，复用函数paddle.linalg.dot