Skip to content
项目
群组
代码片段
帮助
当前项目
正在载入...
登录 / 注册
切换导航面板
P
pytensor
项目
项目
详情
活动
周期分析
仓库
仓库
文件
提交
分支
标签
贡献者
图表
比较
统计图
议题
0
议题
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
CI / CD
CI / CD
流水线
作业
日程
统计图
Wiki
Wiki
代码片段
代码片段
成员
成员
折叠边栏
关闭边栏
活动
图像
聊天
创建新问题
作业
提交
问题看板
Open sidebar
testgroup
pytensor
Commits
42587563
提交
42587563
authored
12月 05, 2025
作者:
Ricardo Vieira
提交者:
Ricardo Vieira
12月 13, 2025
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Numba CAReduce: respect acc_dtype
Also fix infinity identities for unsigned integers
上级
0cc6314b
隐藏空白字符变更
内嵌
并排
正在显示
4 个修改的文件
包含
190 行增加
和
83 行删除
+190
-83
elemwise.py
pytensor/link/numba/dispatch/elemwise.py
+142
-80
elemwise.py
pytensor/tensor/elemwise.py
+4
-1
math.py
pytensor/tensor/math.py
+4
-1
test_elemwise.py
tests/link/numba/test_elemwise.py
+40
-1
没有找到文件。
pytensor/link/numba/dispatch/elemwise.py
浏览文件 @
42587563
...
@@ -2,7 +2,6 @@ from functools import singledispatch
...
@@ -2,7 +2,6 @@ from functools import singledispatch
from
hashlib
import
sha256
from
hashlib
import
sha256
from
textwrap
import
dedent
,
indent
from
textwrap
import
dedent
,
indent
import
numba
import
numpy
as
np
import
numpy
as
np
from
numba.core.extending
import
overload
from
numba.core.extending
import
overload
from
numpy.lib.array_utils
import
normalize_axis_index
,
normalize_axis_tuple
from
numpy.lib.array_utils
import
normalize_axis_index
,
normalize_axis_tuple
...
@@ -15,6 +14,7 @@ from pytensor.link.numba.cache import (
...
@@ -15,6 +14,7 @@ from pytensor.link.numba.cache import (
)
)
from
pytensor.link.numba.dispatch
import
basic
as
numba_basic
from
pytensor.link.numba.dispatch
import
basic
as
numba_basic
from
pytensor.link.numba.dispatch.basic
import
(
from
pytensor.link.numba.dispatch.basic
import
(
create_tuple_string
,
numba_funcify_and_cache_key
,
numba_funcify_and_cache_key
,
register_funcify_and_cache_key
,
register_funcify_and_cache_key
,
register_funcify_default_op_cache_key
,
register_funcify_default_op_cache_key
,
...
@@ -126,10 +126,12 @@ if {res}[{idx}] > {arr}:
...
@@ -126,10 +126,12 @@ if {res}[{idx}] > {arr}:
def
create_multiaxis_reducer
(
def
create_multiaxis_reducer
(
scalar_op
,
scalar_op
,
*
,
identity
,
identity
,
axes
,
axes
,
ndim
,
ndim
,
dtype
,
acc_dtype
=
None
,
out_dtype
,
keepdims
:
bool
=
False
,
keepdims
:
bool
=
False
,
):
):
r"""Construct a function that reduces multiple axes.
r"""Construct a function that reduces multiple axes.
...
@@ -139,17 +141,46 @@ def create_multiaxis_reducer(
...
@@ -139,17 +141,46 @@ def create_multiaxis_reducer(
.. code-block:: python
.. code-block:: python
def careduce_add(x):
def careduce_add(x):
# For x.ndim == 3 and axes == (0, 1) and scalar_op == "Add"
x_shape = x.shape
x_shape = x.shape
res_shape = x_shape[2]
res_shape = (x_shape[0], x_shape[1])
res = np.full(res_shape, numba_basic.to_scalar(0.0), dtype=out_dtype)
# identity = 0.0
res = np.full(res_shape, identity, dtype=np.float64)
for i0 in range(x_shape[0]):
for i1 in range(x_shape[1]):
for i2 in range(x_shape[2]):
res[i0, i1] += x[i0, i1, i2]
return res
If accumulation dtype differs from output_dtype
.. code-block:: python
def careduce_add(x):
x_shape = x.shape
res_shape = (x_shape[0], x_shape[1])
# identity = 0.0
res = np.full(res_shape, identity, dtype=np.float64)
for i0 in range(x_shape[0]):
for i0 in range(x_shape[0]):
for i1 in range(x_shape[1]):
for i1 in range(x_shape[1]):
for i2 in range(x_shape[2]):
for i2 in range(x_shape[2]):
res[i2] += x[i0, i1, i2]
res[i0, i1] += x[i0, i1, i2]
return res.astype(np.int32)
Full reductions accumulate on scalars
.. code-block:: python
def careduce_mul(x):
x_shape = x.shape
res_shape = ()
# identity = 1.0
res = identity
for i0 in range(x_shape[0]):
for i1 in range(x_shape[1]):
for i2 in range(x_shape[2]):
res *= x[i0, i1, i2]
return np.array(res, dtype=np.int32)
return res
Parameters
Parameters
==========
==========
...
@@ -161,7 +192,9 @@ def create_multiaxis_reducer(
...
@@ -161,7 +192,9 @@ def create_multiaxis_reducer(
The axes to reduce.
The axes to reduce.
ndim:
ndim:
The number of dimensions of the input variable.
The number of dimensions of the input variable.
dtype:
acc_dtype: dtype, optional
The data type used during accumulation. Defaults to out_dtype if not provided
out_dtype:
The data type of the result.
The data type of the result.
keepdims: boolean, default False
keepdims: boolean, default False
Whether to keep the reduced dimensions.
Whether to keep the reduced dimensions.
...
@@ -179,19 +212,23 @@ def create_multiaxis_reducer(
...
@@ -179,19 +212,23 @@ def create_multiaxis_reducer(
"Cannot keep multiple dimensions when reducing multiple axes"
"Cannot keep multiple dimensions when reducing multiple axes"
)
)
out_dtype
=
np
.
dtype
(
out_dtype
)
acc_dtype
=
out_dtype
if
acc_dtype
is
None
else
np
.
dtype
(
acc_dtype
)
# Numba doesn't allow converting complex to real with a simple `astype`
complex_to_real
=
acc_dtype
.
kind
==
"c"
and
out_dtype
.
kind
!=
"c"
out_dtype_str
=
f
"np.{out_dtype.name}"
acc_dtype_str
=
f
"np.{acc_dtype.name}"
careduce_fn_name
=
f
"careduce_{scalar_op}"
careduce_fn_name
=
f
"careduce_{scalar_op}"
identity
=
str
(
identity
)
if
acc_dtype
.
kind
in
"ui"
and
not
np
.
isfinite
(
identity
):
if
identity
==
"inf"
:
if
np
.
isposinf
(
identity
):
identity
=
"np.inf"
identity
=
np
.
iinfo
(
acc_dtype
)
.
max
elif
identity
==
"-inf"
:
else
:
identity
=
"-np.inf"
identity
=
np
.
iinfo
(
acc_dtype
)
.
min
global_env
=
{
# Make sure it has the correct dtype
"np"
:
np
,
identity
=
getattr
(
np
,
acc_dtype
.
name
)(
identity
)
"numba_basic"
:
numba_basic
,
"out_dtype"
:
dtype
,
}
complete_reduction
=
len
(
axes
)
==
ndim
complete_reduction
=
len
(
axes
)
==
ndim
kept_axis
=
tuple
(
i
for
i
in
range
(
ndim
)
if
i
not
in
axes
)
kept_axis
=
tuple
(
i
for
i
in
range
(
ndim
)
if
i
not
in
axes
)
...
@@ -209,17 +246,23 @@ def create_multiaxis_reducer(
...
@@ -209,17 +246,23 @@ def create_multiaxis_reducer(
scalar_op
,
res_indices
,
"res"
,
f
"x[{arr_indices}]"
scalar_op
,
res_indices
,
"res"
,
f
"x[{arr_indices}]"
)
)
res_shape
=
f
"({', '.join(f'x_shape[{i}]' for i in kept_axis)})"
res_shape
=
create_tuple_string
([
f
"x_shape[{i}]"
for
i
in
kept_axis
])
if
complete_reduction
and
ndim
>
0
:
if
complete_reduction
and
ndim
>
0
:
# We accumulate on a scalar, not an array
# We accumulate on a scalar, not an array
res_creator
=
f
"np.asarray({identity}).astype(out_dtype).item()
"
res_creator
=
"identity
"
inplace_update_stmt
=
inplace_update_stmt
.
replace
(
"res[()]"
,
"res"
)
inplace_update_stmt
=
inplace_update_stmt
.
replace
(
"res[()]"
,
"res"
)
return_obj
=
"np.asarray(res)"
if
complex_to_real
:
return_obj
=
f
"np.array(res).real.astype({out_dtype_str})"
else
:
return_obj
=
f
"np.array(res, dtype={out_dtype_str})"
else
:
else
:
res_creator
=
(
res_creator
=
f
"np.full(res_shape, identity, dtype={acc_dtype_str})"
f
"np.full({res_shape}, np.asarray({identity}).item(), dtype=out_dtype)"
if
complex_to_real
:
)
return_obj
=
f
"res.real.astype({out_dtype_str})"
return_obj
=
"res"
else
:
return_obj
=
(
"res"
if
out_dtype
==
acc_dtype
else
f
"res.astype({out_dtype_str})"
)
if
keepdims
:
if
keepdims
:
[
axis
]
=
axes
[
axis
]
=
axes
...
@@ -230,6 +273,7 @@ def create_multiaxis_reducer(
...
@@ -230,6 +273,7 @@ def create_multiaxis_reducer(
def {careduce_fn_name}(x):
def {careduce_fn_name}(x):
x_shape = x.shape
x_shape = x.shape
res_shape = {res_shape}
res_shape = {res_shape}
# identity = {identity}
res = {res_creator}
res = {res_creator}
"""
"""
)
)
...
@@ -239,13 +283,12 @@ def create_multiaxis_reducer(
...
@@ -239,13 +283,12 @@ def create_multiaxis_reducer(
" "
*
(
4
+
4
*
axis
),
" "
*
(
4
+
4
*
axis
),
)
)
careduce_def_src
+=
indent
(
inplace_update_stmt
,
" "
*
(
4
+
4
*
ndim
))
careduce_def_src
+=
indent
(
inplace_update_stmt
,
" "
*
(
4
+
4
*
ndim
))
careduce_def_src
+=
"
\n
\n
"
careduce_def_src
+=
"
\n
"
careduce_def_src
+=
indent
(
f
"return {return_obj}"
,
" "
*
4
)
careduce_def_src
+=
indent
(
f
"return {return_obj}"
,
" "
*
4
)
careduce_fn
=
compile_numba_function_src
(
careduce_fn
=
compile_numba_function_src
(
careduce_def_src
,
careduce_fn_name
,
{
**
globals
(),
**
global_env
}
careduce_def_src
,
careduce_fn_name
,
globals
()
|
{
"np"
:
np
,
"identity"
:
identity
}
)
)
return
careduce_fn
return
careduce_fn
...
@@ -356,24 +399,18 @@ def numba_funcify_CAReduce(op, node, **kwargs):
...
@@ -356,24 +399,18 @@ def numba_funcify_CAReduce(op, node, **kwargs):
acc_dtype
=
op
.
acc_dtype
acc_dtype
=
op
.
acc_dtype
else
:
else
:
acc_dtype
=
node
.
outputs
[
0
]
.
type
.
dtype
acc_dtype
=
node
.
outputs
[
0
]
.
type
.
dtype
np_acc_dtype
=
np
.
dtype
(
acc_dtype
)
scalar_op_identity
=
op
.
scalar_op
.
identity
if
np_acc_dtype
.
kind
==
"i"
and
not
np
.
isfinite
(
scalar_op_identity
):
if
np
.
isposinf
(
scalar_op_identity
):
scalar_op_identity
=
np
.
iinfo
(
np_acc_dtype
)
.
max
else
:
scalar_op_identity
=
np
.
iinfo
(
np_acc_dtype
)
.
min
# Make sure it has the correct dtype
scalar_op_identity
=
np
.
array
(
scalar_op_identity
,
dtype
=
np_acc_dtype
)
out_dtype
=
np
.
dtype
(
node
.
outputs
[
0
]
.
type
.
dtype
)
out_dtype
=
np
.
dtype
(
node
.
outputs
[
0
]
.
type
.
dtype
)
if
isinstance
(
op
,
Sum
)
and
node
.
inputs
[
0
]
.
ndim
==
len
(
axes
):
if
(
isinstance
(
op
,
Sum
)
and
node
.
inputs
[
0
]
.
ndim
==
len
(
axes
)
and
out_dtype
==
acc_dtype
):
# Slightly faster for this case
# Slightly faster for this case
@numba_basic.numba_njit
@numba_basic.numba_njit
def
impl_sum
(
array
):
def
impl_sum
(
array
):
return
np
.
a
sarray
(
array
.
sum
(),
dtype
=
np_acc_dtype
)
.
astype
(
out_dtype
)
return
np
.
a
rray
(
array
.
sum
()
)
careduce_fn
=
impl_sum
# Some tests look for this name
careduce_fn
=
impl_sum
# Some tests look for this name
...
@@ -381,16 +418,26 @@ def numba_funcify_CAReduce(op, node, **kwargs):
...
@@ -381,16 +418,26 @@ def numba_funcify_CAReduce(op, node, **kwargs):
ndim
=
node
.
inputs
[
0
]
.
ndim
ndim
=
node
.
inputs
[
0
]
.
ndim
careduce_py_fn
=
create_multiaxis_reducer
(
careduce_py_fn
=
create_multiaxis_reducer
(
op
.
scalar_op
,
op
.
scalar_op
,
scalar_op_identity
,
identity
=
op
.
scalar_op
.
identity
,
axes
,
axes
=
axes
,
ndim
,
ndim
=
ndim
,
out_dtype
,
acc_dtype
=
acc_dtype
,
out_dtype
=
out_dtype
,
)
)
careduce_fn
=
numba_basic
.
numba_njit
(
careduce_py_fn
,
boundscheck
=
False
)
careduce_fn
=
numba_basic
.
numba_njit
(
careduce_py_fn
,
boundscheck
=
False
)
cache_version
=
1
careduce_key
=
sha256
(
careduce_key
=
sha256
(
str
(
str
(
(
type
(
op
),
type
(
op
.
scalar_op
),
axes
,
acc_dtype
,
scalar_op_identity
.
item
())
(
type
(
op
),
type
(
op
.
scalar_op
),
axes
,
out_dtype
,
acc_dtype
,
op
.
scalar_op
.
identity
,
cache_version
,
)
)
.
encode
()
)
.
encode
()
)
.
hexdigest
()
)
.
hexdigest
()
return
careduce_fn
,
careduce_key
return
careduce_fn
,
careduce_key
...
@@ -449,18 +496,26 @@ def numba_funcify_DimShuffle(op: DimShuffle, node, **kwargs):
...
@@ -449,18 +496,26 @@ def numba_funcify_DimShuffle(op: DimShuffle, node, **kwargs):
@register_funcify_default_op_cache_key
(
Softmax
)
@register_funcify_default_op_cache_key
(
Softmax
)
def
numba_funcify_Softmax
(
op
,
node
,
**
kwargs
):
def
numba_funcify_Softmax
(
op
,
node
,
**
kwargs
):
x_at
=
node
.
inputs
[
0
]
ndim
=
node
.
inputs
[
0
]
.
type
.
ndim
x_dtype
=
x_at
.
type
.
numpy_dtype
inp_dtype
=
node
.
inputs
[
0
]
.
type
.
numpy_dtype
x_dtype
=
numba
.
np
.
numpy_support
.
from_dtype
(
x_dtype
)
axis
=
op
.
axis
axis
=
op
.
axis
if
axis
is
not
None
:
if
ndim
>
1
and
axis
is
not
None
:
axis
=
normalize_axis_index
(
axis
,
x_at
.
ndim
)
reduce_max_py
=
create_multiaxis_reducer
(
reduce_max_py
=
create_multiaxis_reducer
(
maximum
,
-
np
.
inf
,
axis
,
x_at
.
ndim
,
x_dtype
,
keepdims
=
True
maximum
,
identity
=-
np
.
inf
,
axes
=
(
axis
,),
ndim
=
ndim
,
out_dtype
=
inp_dtype
,
keepdims
=
True
,
)
)
reduce_sum_py
=
create_multiaxis_reducer
(
reduce_sum_py
=
create_multiaxis_reducer
(
add_as
,
0.0
,
(
axis
,),
x_at
.
ndim
,
x_dtype
,
keepdims
=
True
add_as
,
identity
=
0.0
,
axes
=
(
axis
,),
ndim
=
ndim
,
out_dtype
=
inp_dtype
,
keepdims
=
True
,
)
)
jit_fn
=
numba_basic
.
numba_njit
(
boundscheck
=
False
)
jit_fn
=
numba_basic
.
numba_njit
(
boundscheck
=
False
)
...
@@ -470,29 +525,32 @@ def numba_funcify_Softmax(op, node, **kwargs):
...
@@ -470,29 +525,32 @@ def numba_funcify_Softmax(op, node, **kwargs):
reduce_max
=
np
.
max
reduce_max
=
np
.
max
reduce_sum
=
np
.
sum
reduce_sum
=
np
.
sum
def
softmax_py_fn
(
x
):
@numba_basic.numba_njit
(
boundscheck
=
False
)
def
softmax
(
x
):
z
=
reduce_max
(
x
)
z
=
reduce_max
(
x
)
e_x
=
np
.
exp
(
x
-
z
)
e_x
=
np
.
exp
(
x
-
z
)
w
=
reduce_sum
(
e_x
)
w
=
reduce_sum
(
e_x
)
sm
=
e_x
/
w
sm
=
e_x
/
w
return
sm
return
sm
softmax
=
numba_basic
.
numba_njit
(
softmax_py_fn
,
boundscheck
=
False
)
cache_version
=
1
return
softmax
,
cache_version
return
softmax
@register_funcify_default_op_cache_key
(
SoftmaxGrad
)
@register_funcify_default_op_cache_key
(
SoftmaxGrad
)
def
numba_funcify_SoftmaxGrad
(
op
,
node
,
**
kwargs
):
def
numba_funcify_SoftmaxGrad
(
op
,
node
,
**
kwargs
):
sm_at
=
node
.
inputs
[
1
]
ndim
=
node
.
inputs
[
0
]
.
type
.
ndim
sm_dtype
=
sm_at
.
type
.
numpy_dtype
inp_dtype
=
node
.
inputs
[
0
]
.
type
.
numpy_dtype
sm_dtype
=
numba
.
np
.
numpy_support
.
from_dtype
(
sm_dtype
)
axis
=
op
.
axis
axis
=
op
.
axis
if
axis
is
not
None
:
if
ndim
>
1
and
axis
is
not
None
:
axis
=
normalize_axis_index
(
axis
,
sm_at
.
ndim
)
reduce_sum_py
=
create_multiaxis_reducer
(
reduce_sum_py
=
create_multiaxis_reducer
(
add_as
,
0.0
,
(
axis
,),
sm_at
.
ndim
,
sm_dtype
,
keepdims
=
True
add_as
,
identity
=
0.0
,
axes
=
(
axis
,),
ndim
=
ndim
,
out_dtype
=
inp_dtype
,
keepdims
=
True
,
)
)
jit_fn
=
numba_basic
.
numba_njit
(
boundscheck
=
False
)
jit_fn
=
numba_basic
.
numba_njit
(
boundscheck
=
False
)
...
@@ -500,36 +558,39 @@ def numba_funcify_SoftmaxGrad(op, node, **kwargs):
...
@@ -500,36 +558,39 @@ def numba_funcify_SoftmaxGrad(op, node, **kwargs):
else
:
else
:
reduce_sum
=
np
.
sum
reduce_sum
=
np
.
sum
def
softmax_grad_py_fn
(
dy
,
sm
):
@numba_basic.numba_njit
(
boundscheck
=
False
)
def
softmax_grad
(
dy
,
sm
):
dy_times_sm
=
dy
*
sm
dy_times_sm
=
dy
*
sm
sum_dy_times_sm
=
reduce_sum
(
dy_times_sm
)
sum_dy_times_sm
=
reduce_sum
(
dy_times_sm
)
dx
=
dy_times_sm
-
sum_dy_times_sm
*
sm
dx
=
dy_times_sm
-
sum_dy_times_sm
*
sm
return
dx
return
dx
softmax_grad
=
numba_basic
.
numba_njit
(
softmax_grad_py_fn
,
boundscheck
=
False
)
cache_version
=
1
return
softmax_grad
,
cache_version
return
softmax_grad
@register_funcify_default_op_cache_key
(
LogSoftmax
)
@register_funcify_default_op_cache_key
(
LogSoftmax
)
def
numba_funcify_LogSoftmax
(
op
,
node
,
**
kwargs
):
def
numba_funcify_LogSoftmax
(
op
,
node
,
**
kwargs
):
x_at
=
node
.
inputs
[
0
]
ndim
=
node
.
inputs
[
0
]
.
type
.
ndim
x_dtype
=
x_at
.
type
.
numpy_dtype
inp_dtype
=
node
.
inputs
[
0
]
.
type
.
numpy_dtype
x_dtype
=
numba
.
np
.
numpy_support
.
from_dtype
(
x_dtype
)
axis
=
op
.
axis
axis
=
op
.
axis
if
axis
is
not
None
:
if
ndim
>
1
and
axis
is
not
None
:
axis
=
normalize_axis_index
(
axis
,
x_at
.
ndim
)
reduce_max_py
=
create_multiaxis_reducer
(
reduce_max_py
=
create_multiaxis_reducer
(
maximum
,
maximum
,
-
np
.
inf
,
identity
=
-
np
.
inf
,
(
axis
,),
axes
=
(
axis
,),
x_at
.
ndim
,
ndim
=
ndim
,
x
_dtype
,
out_dtype
=
inp
_dtype
,
keepdims
=
True
,
keepdims
=
True
,
)
)
reduce_sum_py
=
create_multiaxis_reducer
(
reduce_sum_py
=
create_multiaxis_reducer
(
add_as
,
0.0
,
(
axis
,),
x_at
.
ndim
,
x_dtype
,
keepdims
=
True
add_as
,
identity
=
0.0
,
axes
=
(
axis
,),
ndim
=
ndim
,
out_dtype
=
inp_dtype
,
keepdims
=
True
,
)
)
jit_fn
=
numba_basic
.
numba_njit
(
boundscheck
=
False
)
jit_fn
=
numba_basic
.
numba_njit
(
boundscheck
=
False
)
...
@@ -539,13 +600,14 @@ def numba_funcify_LogSoftmax(op, node, **kwargs):
...
@@ -539,13 +600,14 @@ def numba_funcify_LogSoftmax(op, node, **kwargs):
reduce_max
=
np
.
max
reduce_max
=
np
.
max
reduce_sum
=
np
.
sum
reduce_sum
=
np
.
sum
def
log_softmax_py_fn
(
x
):
@numba_basic.numba_njit
(
boundscheck
=
False
)
def
log_softmax
(
x
):
xdev
=
x
-
reduce_max
(
x
)
xdev
=
x
-
reduce_max
(
x
)
lsm
=
xdev
-
np
.
log
(
reduce_sum
(
np
.
exp
(
xdev
)))
lsm
=
xdev
-
np
.
log
(
reduce_sum
(
np
.
exp
(
xdev
)))
return
lsm
return
lsm
log_softmax
=
numba_basic
.
numba_njit
(
log_softmax_py_fn
,
boundscheck
=
False
)
cache_version
=
1
return
log_softmax
return
log_softmax
,
cache_version
@register_funcify_default_op_cache_key
(
Argmax
)
@register_funcify_default_op_cache_key
(
Argmax
)
...
...
pytensor/tensor/elemwise.py
浏览文件 @
42587563
...
@@ -1391,7 +1391,10 @@ class CAReduce(COp):
...
@@ -1391,7 +1391,10 @@ class CAReduce(COp):
return
f
"axes={list(axis)}"
return
f
"axes={list(axis)}"
def
__str__
(
self
):
def
__str__
(
self
):
return
f
"{type(self).__name__}{{{self.scalar_op}, {self._axis_str()}}}"
if
self
.
acc_dtype
!=
self
.
dtype
:
return
f
"{type(self).__name__}{{{self.scalar_op}, {self._axis_str()}, acc={self.acc_dtype}}}"
else
:
return
f
"{type(self).__name__}{{{self.scalar_op}, {self._axis_str()}}}"
def
perform
(
self
,
node
,
inp
,
out
):
def
perform
(
self
,
node
,
inp
,
out
):
(
input
,)
=
inp
(
input
,)
=
inp
...
...
pytensor/tensor/math.py
浏览文件 @
42587563
...
@@ -357,7 +357,10 @@ def max_and_argmax(a, axis=None, keepdims=False):
...
@@ -357,7 +357,10 @@ def max_and_argmax(a, axis=None, keepdims=False):
class
FixedOpCAReduce
(
CAReduce
):
class
FixedOpCAReduce
(
CAReduce
):
def
__str__
(
self
):
def
__str__
(
self
):
return
f
"{type(self).__name__}{{{self._axis_str()}}}"
if
self
.
dtype
!=
self
.
acc_dtype
:
return
f
"{type(self).__name__}{{{self._axis_str()}, acc={self.acc_dtype}}}"
else
:
return
f
"{type(self).__name__}{{{self._axis_str()}}}"
class
NonZeroDimsCAReduce
(
FixedOpCAReduce
):
class
NonZeroDimsCAReduce
(
FixedOpCAReduce
):
...
...
tests/link/numba/test_elemwise.py
浏览文件 @
42587563
...
@@ -13,7 +13,7 @@ from pytensor.compile.ops import deep_copy_op
...
@@ -13,7 +13,7 @@ from pytensor.compile.ops import deep_copy_op
from
pytensor.gradient
import
grad
from
pytensor.gradient
import
grad
from
pytensor.scalar
import
Composite
,
float64
from
pytensor.scalar
import
Composite
,
float64
from
pytensor.scalar
import
add
as
scalar_add
from
pytensor.scalar
import
add
as
scalar_add
from
pytensor.tensor
import
blas
,
tensor
from
pytensor.tensor
import
blas
,
matrix
,
tensor
,
tensor3
from
pytensor.tensor.elemwise
import
CAReduce
,
DimShuffle
,
Elemwise
from
pytensor.tensor.elemwise
import
CAReduce
,
DimShuffle
,
Elemwise
from
pytensor.tensor.math
import
All
,
Any
,
Max
,
Min
,
Prod
,
ProdWithoutZeros
,
Sum
from
pytensor.tensor.math
import
All
,
Any
,
Max
,
Min
,
Prod
,
ProdWithoutZeros
,
Sum
from
pytensor.tensor.special
import
LogSoftmax
,
Softmax
,
SoftmaxGrad
from
pytensor.tensor.special
import
LogSoftmax
,
Softmax
,
SoftmaxGrad
...
@@ -366,6 +366,45 @@ def test_CAReduce(careduce_fn, axis, v):
...
@@ -366,6 +366,45 @@ def test_CAReduce(careduce_fn, axis, v):
assert
isinstance
(
node
.
op
,
CAReduce
)
assert
isinstance
(
node
.
op
,
CAReduce
)
@pytest.mark.parametrize
(
"axis"
,
(
-
1
,
(
0
,
-
1
),
None
))
def
test_CAReduce_respects_acc_dtype
(
axis
):
x
=
tensor3
(
"x"
,
dtype
=
"int8"
)
out
=
x
.
sum
(
dtype
=
"int8"
,
acc_dtype
=
"int64"
,
axis
=
axis
)
# Choose values that would overflow if accumulated internally in int8
max_int8
=
np
.
iinfo
(
np
.
int8
)
.
max
test_x
=
np
.
array
([
max_int8
,
5
,
max_int8
,
-
max_int8
,
5
,
-
max_int8
],
dtype
=
np
.
int8
)
test_x
=
np
.
broadcast_to
(
test_x
,
(
6
,
2
,
6
))
.
copy
()
_
,
[
res
]
=
compare_numba_and_py
(
[
x
],
[
out
],
[
test_x
],
)
if
axis
==
-
1
:
assert
np
.
all
(
res
==
10
)
elif
axis
==
(
0
,
-
1
):
assert
np
.
all
(
res
==
60
)
elif
axis
is
None
:
assert
res
==
120
@pytest.mark.parametrize
(
"axis"
,
(
1
,
None
))
def
test_CAReduce_acc_complex_out_float
(
axis
):
x
=
matrix
(
"x"
,
dtype
=
"complex128"
)
out
=
x
.
sum
(
dtype
=
"float64"
,
axis
=
axis
)
test_x
=
np
.
array
([[
1
+
0.5
j
,
2
-
0.5
j
],
[
3
+
0.5
j
,
4
-
0.5
j
]],
dtype
=
"complex128"
)
compare_numba_and_py
([
x
],
[
out
],
[
test_x
])
@pytest.mark.parametrize
(
"axis"
,
(
-
1
,
(
0
,
-
1
),
None
))
def
test_CAReduce_discrete_infinity_identity
(
axis
):
rng
=
np
.
random
.
default_rng
(
337
)
x
=
tensor3
(
"x"
,
dtype
=
"int8"
)
out
=
x
.
max
(
axis
)
compare_numba_and_py
(
[
x
],
[
out
],
[
rng
.
integers
(
-
127
,
127
,
size
=
(
6
,
6
,
6
))
.
astype
(
"int8"
)]
)
def
test_scalar_Elemwise_Clip
():
def
test_scalar_Elemwise_Clip
():
a
=
pt
.
scalar
(
"a"
)
a
=
pt
.
scalar
(
"a"
)
b
=
pt
.
scalar
(
"b"
)
b
=
pt
.
scalar
(
"b"
)
...
...
编写
预览
Markdown
格式
0%
重试
或
添加新文件
添加附件
取消
您添加了
0
人
到此讨论。请谨慎行事。
请先完成此评论的编辑!
取消
请
注册
或者
登录
后发表评论