diff --git a/oneflow/python/nn/modules/masked_fill.py b/oneflow/python/nn/modules/masked_fill.py
index 28b00766ac7c97a45b84aef091611017f8f6e0a2..19701c42a0f4ac4aad83186772a2332224238f61 100644
--- a/oneflow/python/nn/modules/masked_fill.py
+++ b/oneflow/python/nn/modules/masked_fill.py
@@ -34,7 +34,7 @@ class MaskedFill(Module):
@oneflow_export("masked_fill")
@register_tensor_op("masked_fill")
@experimental_api
-def masked_fill_op(tensor, mask, value):
+def masked_fill_op(input, mask, value):
r"""
Fills elements of :attr:`self` tensor with :attr:`value` where :attr:`mask` is True.
The shape of :attr:`mask` must be broadcastable with the shape of the underlying tensor.
@@ -72,7 +72,7 @@ def masked_fill_op(tensor, mask, value):
# [-1.9009, 8.7654, 8.7654, 8.7654]]], dtype=oneflow.float32)
"""
- return MaskedFill(value)(tensor, mask)
+ return MaskedFill(value)(input, mask)
if __name__ == "__main__":
diff --git a/oneflow/python/nn/modules/math_ops.py b/oneflow/python/nn/modules/math_ops.py
index 0d8296a199247e67dbaa0f2080528f5ae97ef9ad..b7ad7edeb4539c9a1f013ee6ef42094080d05b0b 100644
--- a/oneflow/python/nn/modules/math_ops.py
+++ b/oneflow/python/nn/modules/math_ops.py
@@ -66,13 +66,13 @@ class BroadcastMul(Module):
@oneflow_export("mul")
@register_tensor_op("mul")
@experimental_api
-def _mul(x, y):
- r"""Computes the multiplication of x by y for each element, scalar and broadcast promotation are supported.
+def _mul(input, other):
+ r"""Computes the multiplication of input by other for each element, scalar and broadcast promotation are supported.
The formula is:
.. math::
- out = x \times y
+ out = input \times other
For example:
@@ -83,40 +83,40 @@ def _mul(x, y):
>>> flow.enable_eager_execution()
# element-wise multiply
- >>> x = flow.Tensor(np.random.randn(2,3))
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.mul(x,y).numpy()
+ >>> input = flow.Tensor(np.random.randn(2,3))
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.mul(input,other).numpy()
>>> out.shape
(2, 3)
# scalar mutiply
- >>> x = 5
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.mul(x,y).numpy()
+ >>> input = 5
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.mul(input,other).numpy()
>>> out.shape
(2, 3)
# broadcast mutiply
- >>> x = flow.Tensor(np.random.randn(1,1))
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.mul(x,y).numpy()
+ >>> input = flow.Tensor(np.random.randn(1,1))
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.mul(input,other).numpy()
>>> out.shape
(2, 3)
"""
- if isinstance(x, (int, float)):
- return ScalarMul(x)(y)
- elif isinstance(y, (int, float)):
- return ScalarMul(y)(x)
- elif x.shape == y.shape:
- return ElementwiseMul()(x, y)
- elif x.shape == (1,):
- return ScalarMulByTensor()(y, x)
- elif y.shape == (1,):
- return ScalarMulByTensor()(x, y)
+ if isinstance(input, (int, float)):
+ return ScalarMul(input)(other)
+ elif isinstance(other, (int, float)):
+ return ScalarMul(other)(input)
+ elif input.shape == other.shape:
+ return ElementwiseMul()(input, other)
+ elif input.shape == (1,):
+ return ScalarMulByTensor()(other, input)
+ elif other.shape == (1,):
+ return ScalarMulByTensor()(input, other)
else:
- return BroadcastMul()(x, y)
+ return BroadcastMul()(input, other)
class Variance(Module):
@@ -207,12 +207,12 @@ class ScalarAdd(Module):
@oneflow_export("sub")
@register_tensor_op("sub")
@experimental_api
-def _sub(x, y):
- r"""Computes the subtraction of x by y for each element, scalar and broadcast promotation are supported.
+def _sub(input, other):
+ r"""Computes the subtraction of input by other for each element, scalar and broadcast promotation are supported.
The formula is:
.. math::
- out = x - y
+ out = input - other
For example:
@@ -223,39 +223,39 @@ def _sub(x, y):
>>> flow.enable_eager_execution()
# element-wise subtract
- >>> x = flow.Tensor(np.random.randn(2,3))
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.sub(x,y).numpy()
+ >>> input = flow.Tensor(np.random.randn(2,3))
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.sub(input,other).numpy()
>>> out.shape
(2, 3)
# scalar subtract
- >>> x = 5
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.sub(x,y).numpy()
+ >>> input = 5
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.sub(input,other).numpy()
>>> out.shape
(2, 3)
# broadcast subtract
- >>> x = flow.Tensor(np.random.randn(1,1))
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.sub(x,y).numpy()
+ >>> input = flow.Tensor(np.random.randn(1,1))
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.sub(input,other).numpy()
>>> out.shape
(2, 3)
"""
- if isinstance(x, (int, float)):
- return ScalarAdd(x)(ScalarMul(-1)(y))
- elif isinstance(y, (int, float)):
- return ScalarAdd(-1 * y)(x)
- elif x.shape == y.shape:
+ if isinstance(input, (int, float)):
+ return ScalarAdd(input)(ScalarMul(-1)(other))
+ elif isinstance(other, (int, float)):
+ return ScalarAdd(-1 * other)(input)
+ elif input.shape == other.shape:
# TODO: add element-wise op
- return BroadcastSub()(x, y)
- elif y.shape == (1,):
- return ScalarSubByTensor()(x, y)
+ return BroadcastSub()(input, other)
+ elif other.shape == (1,):
+ return ScalarSubByTensor()(input, other)
else:
- return BroadcastSub()(x, y)
+ return BroadcastSub()(input, other)
class BroadcastDiv(Module):
@@ -277,16 +277,16 @@ class ScalarDivByTensor(Module):
@oneflow_export("div")
@register_tensor_op("div")
@experimental_api
-def _div(x, y):
- r"""Computes the division of x by y for each element, scalar and broadcast promotation are supported.
+def _div(input, other):
+ r"""Computes the division of input by other for each element, scalar and broadcast promotation are supported.
The formula is:
.. math::
- out = \frac{X}{Y}
+ out = \frac{input}{other}
Args:
- x (Union[int, float, flow.Tensor]): X.
- y (Union[int, float, flow.Tensor]): Y.
+ input (Union[int, float, flow.Tensor]): input.
+ other (Union[int, float, flow.Tensor]): other.
For example:
@@ -297,42 +297,42 @@ def _div(x, y):
>>> flow.enable_eager_execution()
# element-wise divide
- >>> x = flow.Tensor(np.random.randn(2,3))
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.div(x,y).numpy()
+ >>> input = flow.Tensor(np.random.randn(2,3))
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.div(input,other).numpy()
>>> out.shape
(2, 3)
# scalar divide
- >>> x = 5
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.div(x,y).numpy()
+ >>> input = 5
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.div(input,other).numpy()
>>> out.shape
(2, 3)
# broadcast divide
- >>> x = flow.Tensor(np.random.randn(1,1))
- >>> y = flow.Tensor(np.random.randn(2,3))
- >>> out = flow.div(x,y).numpy()
+ >>> input = flow.Tensor(np.random.randn(1,1))
+ >>> other = flow.Tensor(np.random.randn(2,3))
+ >>> out = flow.div(input,other).numpy()
>>> out.shape
(2, 3)
"""
- if isinstance(x, (int, float)):
- return ScalarMul(x)(flow.experimental.reciprocal(y))
- elif isinstance(y, (int, float)):
- if y == 0 or y == 0.0:
- y = 0.0
+ if isinstance(input, (int, float)):
+ return ScalarMul(input)(flow.experimental.reciprocal(other))
+ elif isinstance(other, (int, float)):
+ if other == 0 or other == 0.0:
+ other = 0.0
else:
- y = 1.0 / (float(y))
- return ScalarMul(y)(x)
- elif x.shape == y.shape:
- return BroadcastDiv()(x, y)
- elif y.shape == (1,):
- return ScalarDivByTensor()(x, y)
+ other = 1.0 / (float(other))
+ return ScalarMul(other)(input)
+ elif input.shape == other.shape:
+ return BroadcastDiv()(input, other)
+ elif other.shape == (1,):
+ return ScalarDivByTensor()(input, other)
else:
- return BroadcastDiv()(x, y)
+ return BroadcastDiv()(input, other)
class Reciprocal(Module):
diff --git a/oneflow/python/test/modules/test_div.py b/oneflow/python/test/modules/test_div.py
index ac39f0e196b1f5172faa4e121f4ec5310a6f3b81..078ddc9a780018d76f3f0af36cc6833bc7e4e7cd 100644
--- a/oneflow/python/test/modules/test_div.py
+++ b/oneflow/python/test/modules/test_div.py
@@ -20,6 +20,7 @@ import numpy as np
import oneflow.experimental as flow
from test_util import GenArgList
+from automated_test_util import *
def _test_div_impl(test_case, shape, device):
@@ -78,6 +79,34 @@ class TestDiv(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
_test_div_impl(test_case, *arg)
+ def test_sub_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ arg_dict["op"] = ["div"]
+ for arg in GenArgList(arg_dict):
+ arg[0](
+ test_case,
+ arg[2],
+ extra_annotations={"other": flow.Tensor},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=2, dim1=3),
+ "other": random_tensor(ndim=2, dim0=2, dim1=3),
+ },
+ device=arg[1],
+ )
+
+ arg[0](
+ test_case,
+ arg[2],
+ extra_annotations={"other": float},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=2, dim1=3),
+ "other": random(0, 5),
+ },
+ device=arg[1],
+ )
+
if __name__ == "__main__":
unittest.main()
diff --git a/oneflow/python/test/modules/test_masked_fill.py b/oneflow/python/test/modules/test_masked_fill.py
index 4142b4fb36f0aa2c6dcc72de1e211c41478ea1ce..82206fd8765ef7db473f6f2048dc1ce2aea89aca 100644
--- a/oneflow/python/test/modules/test_masked_fill.py
+++ b/oneflow/python/test/modules/test_masked_fill.py
@@ -14,85 +14,11 @@ See the License for the specific language governing permissions and
limitations under the License.
"""
import unittest
-from collections import OrderedDict
import numpy as np
import oneflow.experimental as flow
-from test_util import GenArgList
-
-
-def _test_masked_fill(test_case, device):
- input_arr = np.array(
- [
- [
- [-0.13169311, 0.97277078, 1.23305363, 1.56752789],
- [-1.51954275, 1.87629473, -0.53301206, 0.53006478],
- [-1.38244183, -2.63448052, 1.30845795, -0.67144869],
- ],
- [
- [0.41502161, 0.14452418, 0.38968, -1.76905653],
- [0.34675095, -0.7050969, -0.7647731, -0.73233418],
- [-1.90089858, 0.01262963, 0.74693893, 0.57132389],
- ],
- ]
- )
-
- output = np.array(
- [
- [
- [-0.1316931, 8.7654321, 8.7654321, 8.7654321],
- [-1.5195428, 8.7654321, -0.5330121, 8.7654321],
- [-1.3824418, -2.6344805, 8.7654321, -0.6714487],
- ],
- [
- [8.7654321, 8.7654321, 8.7654321, -1.7690565],
- [8.7654321, -0.7050969, -0.7647731, -0.7323342],
- [-1.9008986, 8.7654321, 8.7654321, 8.7654321],
- ],
- ]
- )
-
- fill_value = 8.7654321 # random value e.g. -1e9 3.14
-
- input = flow.Tensor(input_arr, dtype=flow.float32, device=flow.device(device))
- mask = flow.Tensor(
- (input_arr > 0).astype(np.int8), dtype=flow.int, device=flow.device(device)
- )
- of_out = flow.masked_fill(input, mask, value=fill_value)
- test_case.assertTrue(np.allclose(of_out.numpy(), output))
-
-
-def _test_masked_fill_backward(test_case, device):
- input_arr = np.array(
- [
- [
- [-0.13169311, 0.97277078, 1.23305363, 1.56752789],
- [-1.51954275, 1.87629473, -0.53301206, 0.53006478],
- [-1.38244183, -2.63448052, 1.30845795, -0.67144869],
- ],
- [
- [0.41502161, 0.14452418, 0.38968, -1.76905653],
- [0.34675095, -0.7050969, -0.7647731, -0.73233418],
- [-1.90089858, 0.01262963, 0.74693893, 0.57132389],
- ],
- ]
- )
-
- fill_value = -3.1415 # random value e.g. -1e9 3.14
-
- x = flow.Tensor(
- input_arr, dtype=flow.float32, device=flow.device(device), requires_grad=True
- )
- mask = flow.Tensor(
- (input_arr > 0).astype(np.int8), dtype=flow.int, device=flow.device(device)
- )
- y = flow.masked_fill(x, mask, value=fill_value)
- z = y.sum()
- z.backward()
- test_case.assertTrue(
- np.array_equal(x.grad.numpy(), (input_arr < 0).astype(np.float32))
- )
+from automated_test_util import *
@unittest.skipIf(
@@ -100,12 +26,45 @@ def _test_masked_fill_backward(test_case, device):
".numpy() doesn't work in lazy mode",
)
class TestMaskedFill(flow.unittest.TestCase):
- def test_masked_fill(test_case):
- arg_dict = OrderedDict()
- arg_dict["test_fun"] = [_test_masked_fill, _test_masked_fill_backward]
- arg_dict["device"] = ["cpu", "cuda"]
- for arg in GenArgList(arg_dict):
- arg[0](test_case, *arg[1:])
+ def test_masked_fill_aginst_pytorch(test_case):
+ import numpy as np
+ import torch
+
+ def mask_tensor(shape):
+ def generator(_):
+ rng = np.random.default_rng()
+ np_arr = rng.integers(low=0, high=2, size=shape)
+ return (
+ flow.Tensor(np_arr, dtype=flow.int8),
+ torch.tensor(np_arr, dtype=torch.bool),
+ )
+
+ return generator
+
+ for device in ["cpu", "cuda"]:
+ test_flow_against_pytorch(
+ test_case,
+ "masked_fill",
+ extra_annotations={"mask": flow.Tensor, "value": float},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=4, dim1=5),
+ "mask": mask_tensor((4, 5)),
+ "value": constant(3.14),
+ },
+ device=device,
+ )
+
+ test_tensor_against_pytorch(
+ test_case,
+ "masked_fill",
+ extra_annotations={"mask": flow.Tensor, "value": float},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=4, dim1=5),
+ "mask": mask_tensor((4, 5)),
+ "value": constant(3.14),
+ },
+ device=device,
+ )
if __name__ == "__main__":
diff --git a/oneflow/python/test/modules/test_mean.py b/oneflow/python/test/modules/test_mean.py
index 0d5e2f7c3537698dbb8983e578000f1f21f16cd6..7b8fe7e05122383644be2acdbc61838208532996 100644
--- a/oneflow/python/test/modules/test_mean.py
+++ b/oneflow/python/test/modules/test_mean.py
@@ -20,6 +20,7 @@ import numpy as np
import oneflow.experimental as flow
from test_util import GenArgList
+from automated_test_util import *
def _test_mean(test_case, shape, device):
@@ -79,6 +80,13 @@ class TestMean(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
arg[0](test_case, *arg[1:])
+ def test_mean_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ for arg in GenArgList(arg_dict):
+ arg[0](test_case, "mean", device=arg[1])
+
if __name__ == "__main__":
unittest.main()
diff --git a/oneflow/python/test/modules/test_mul.py b/oneflow/python/test/modules/test_mul.py
index 3f85b4d709a55475308504b5f6a5df7969534fe8..8fd3372b49f6cc0fd32bb9a1e4f992440ddcfe1e 100644
--- a/oneflow/python/test/modules/test_mul.py
+++ b/oneflow/python/test/modules/test_mul.py
@@ -20,6 +20,7 @@ import numpy as np
import oneflow.experimental as flow
from test_util import GenArgList
+from automated_test_util import *
def _test_mul_impl(test_case, device):
@@ -111,6 +112,34 @@ class TestMulModule(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
arg[0](test_case, *arg[1:])
+ def test_mul_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ arg_dict["op"] = ["mul"]
+ for arg in GenArgList(arg_dict):
+ arg[0](
+ test_case,
+ arg[2],
+ extra_annotations={"other": flow.Tensor},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=2, dim1=3),
+ "other": random_tensor(ndim=2, dim0=2, dim1=3),
+ },
+ device=arg[1],
+ )
+
+ arg[0](
+ test_case,
+ arg[2],
+ extra_annotations={"other": float},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=2, dim1=3),
+ "other": random(0, 5),
+ },
+ device=arg[1],
+ )
+
if __name__ == "__main__":
unittest.main()
diff --git a/oneflow/python/test/modules/test_reduce_ops.py b/oneflow/python/test/modules/test_reduce_ops.py
index ee3088aa0fc72c435f774dd53babd76c58f1b420..6bf965d212697623011449243b9cd7d823403fff 100644
--- a/oneflow/python/test/modules/test_reduce_ops.py
+++ b/oneflow/python/test/modules/test_reduce_ops.py
@@ -20,6 +20,7 @@ import numpy as np
import oneflow.experimental as flow
from test_util import GenArgList
+from automated_test_util import *
def _test_min(test_case, device, shape, dim, keepdims):
@@ -71,7 +72,7 @@ def _test_min_tensor_function(test_case, device, shape, dim, keepdims):
".numpy() doesn't work in lazy mode",
)
class TestMinModule(flow.unittest.TestCase):
- def test_max(test_case):
+ def test_min(test_case):
arg_dict = OrderedDict()
arg_dict["test_fun"] = [_test_min, _test_min_tensor_function]
arg_dict["device"] = ["cpu", "cuda"]
@@ -81,6 +82,13 @@ class TestMinModule(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
arg[0](test_case, *arg[1:])
+ def test_min_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ for arg in GenArgList(arg_dict):
+ arg[0](test_case, "min", device=arg[1])
+
def _test_max(test_case, device, shape, dim, keepdims):
input_arr = np.random.randn(*shape)
@@ -141,6 +149,13 @@ class TestMaxModule(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
arg[0](test_case, *arg[1:])
+ def test_max_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ for arg in GenArgList(arg_dict):
+ arg[0](test_case, "max", device=arg[1])
+
if __name__ == "__main__":
unittest.main()
diff --git a/oneflow/python/test/modules/test_sub.py b/oneflow/python/test/modules/test_sub.py
index 8792e266b2890309ce3fa721ecc05534d71793ad..f2c3ecd6445413affb8e25218502485261d9f26c 100644
--- a/oneflow/python/test/modules/test_sub.py
+++ b/oneflow/python/test/modules/test_sub.py
@@ -20,6 +20,7 @@ import numpy as np
import oneflow.experimental as flow
from test_util import GenArgList
+from automated_test_util import *
def _test_sub_impl(test_case, shape, device):
@@ -88,6 +89,34 @@ class TestSubModule(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
_test_sub_impl(test_case, *arg)
+ def test_sub_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ arg_dict["op"] = ["sub"]
+ for arg in GenArgList(arg_dict):
+ arg[0](
+ test_case,
+ arg[2],
+ extra_annotations={"other": flow.Tensor},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=2, dim1=3),
+ "other": random_tensor(ndim=2, dim0=2, dim1=3),
+ },
+ device=arg[1],
+ )
+
+ arg[0](
+ test_case,
+ arg[2],
+ extra_annotations={"other": float},
+ extra_generators={
+ "input": random_tensor(ndim=2, dim0=2, dim1=3),
+ "other": random(0, 5),
+ },
+ device=arg[1],
+ )
+
if __name__ == "__main__":
unittest.main()
diff --git a/oneflow/python/test/modules/test_sum.py b/oneflow/python/test/modules/test_sum.py
index d8786add97d236200dd6540f3b49ec6bffb6aec1..a82efe65331f52268dcdbc28a08b822943098616 100644
--- a/oneflow/python/test/modules/test_sum.py
+++ b/oneflow/python/test/modules/test_sum.py
@@ -20,6 +20,7 @@ import numpy as np
import oneflow.experimental as flow
from test_util import GenArgList
+from automated_test_util import *
def _test_sum_impl(test_case, device):
@@ -72,6 +73,13 @@ class TestSumModule(flow.unittest.TestCase):
for arg in GenArgList(arg_dict):
_test_sum_impl(test_case, *arg)
+ def test_sum_against_pytorch(test_case):
+ arg_dict = OrderedDict()
+ arg_dict["test_type"] = [test_flow_against_pytorch, test_tensor_against_pytorch]
+ arg_dict["device"] = ["cpu", "cuda"]
+ for arg in GenArgList(arg_dict):
+ arg[0](test_case, "sum", device=arg[1])
+
if __name__ == "__main__":
unittest.main()