liangcanli
ba5e41c0cb
fix getgrad from list and fix error type
2023-02-21 20:11:19 +08:00
chenfei
1637aebf9c
fix bug of dde in self definition cell
2023-02-18 10:21:32 +08:00
yujianfeng
6eb99fd929
Fix dict grad
2023-02-10 10:15:21 +08:00
i-robot
283819a1c9
!48125 Change custom_vjp to inner api
...
Merge pull request !48125 from YuJianfeng/custom_vjp
2023-01-30 06:15:45 +00:00
yujianfeng
77d10800ca
Change custom_vjp to inner api
2023-01-29 10:02:20 +08:00
liangcanli
dc658a002f
fix get_grad docs and add testcases
2023-01-18 15:24:29 +08:00
yujianfeng
b892806440
Clone the used graph when a grad graph calls another
2023-01-10 20:10:06 +08:00
liangcanli
249c721f5a
add index of grad
2022-12-29 19:10:36 +08:00
zjun
9375c56259
Add high order for pynative
...
Signed-off-by: zjun <zhangjun0@huawei.com>
2022-11-29 15:25:06 +08:00
luochao
605c1a8479
new_construct_bprop
...
move expander files
move expander component to mindspore/core
move the bprop expanders to mindspore/frontend
2022-11-27 04:55:01 -05:00
liangzhibo
7707bcee5f
rectificate the st for gradient
2022-11-18 17:04:39 +08:00
huangbingjian
b98fc0021a
Add api mindspore.jit and mindspore.jit_class.
...
1、ms_function will be removed in a future version and replaced with jit.
2、ms_class will be removed in a future version and replaced with
jit_class.
2022-10-21 10:17:44 +08:00
zjun
da1541e523
Fix grad param
...
Signed-off-by: zjun <zhangjun0@huawei.com>
2022-10-15 15:46:08 +08:00
i-robot
1cf78e1d85
!42060 [feature] add jacrev
...
Merge pull request !42060 from chenzhuo/master_jacrev
2022-09-30 01:28:48 +00:00
chenzhuo
f8b2d3ca60
add jacrev
2022-09-29 09:29:37 +08:00
i-robot
8e983d2027
!43084 takedown testcases that are almost never failing as to save time on gate
...
Merge pull request !43084 from yanghaoran/master
2022-09-28 07:49:31 +00:00
i-robot
c2cf1d3393
!42694 Revert modify in PyNative Grad
...
Merge pull request !42694 from huangbingjian/revert_pynative_grad
2022-09-27 15:51:28 +00:00
yanghaoran
164fada9f5
takedown testcases that are almost never failing as to save time on gate
2022-09-27 22:38:48 +08:00
ZhengXuegui
685f0cc830
support vjp to custom bprop function
2022-09-26 20:04:14 +08:00
i-robot
2a2ce2f500
!40939 [feature] add jacfwd
...
Merge pull request !40939 from chenzhuo/master_jacfwd
2022-09-26 09:40:26 +00:00
huangbingjian
664ffa8f5b
Revert modify in PyNative Grad
2022-09-24 12:52:49 +08:00
chenzhuo
bd01dcad29
add jacfwd
2022-09-23 15:00:53 +08:00
i-robot
9ec79c2657
!42184 Return an empty tuple when getting the gradient of constants.
...
Merge pull request !42184 from huangbingjian/grad_empty_tuple
2022-09-22 11:38:19 +00:00
huangbingjian
ba8a72ecae
Return an empty tuple when getting the gradient of constants.
2022-09-16 17:32:09 +08:00
chenzhuo
9aa6bdbaa2
change jvp
2022-09-16 15:31:55 +08:00
i-robot
242e5c819e
!40599 [feature] change vjp
...
Merge pull request !40599 from chenzhuo/master_vjp
2022-09-15 01:41:13 +00:00
i-robot
5046b06831
!41827 Fix GradOpertion in hypermap controlflow
...
Merge pull request !41827 from huangbingjian/grad_hpyermap
2022-09-14 09:43:47 +00:00
chenzhuo
b742f8aa76
modify vjp
2022-09-14 15:08:16 +08:00
huangbingjian
831a113959
Fix GradOpertion in hypermap controlflow
2022-09-13 15:27:50 +08:00
huangbingjian
101574e8ee
Fix gradient when get_by_list is True and weights is None
2022-09-09 14:41:54 +08:00
huangbingjian
161ff44ce9
Fix GradOperation issues.
...
1、Refactor the GradOperation code in PyNative mode.
2、In PyNative mode, when get_by_list=True is set, even if a single weight is specified, all weights will be wrongly involved in grad.
3、In PyNative mode, when the construct code only returns a single weight, such as `return self.w`, the result of the gradient value of self.w should be 1, but it is ().
4、Optimize the error message of GradOperation/grad in Graph mode.
5、Add testcases for GradOperation/grad.
2022-09-08 10:12:40 +08:00
yujianfeng
06510b0649
Allow tensor to be set const for network argument
2022-08-29 21:12:03 +08:00
i-robot
de4b5b238e
!40686 [bug] fix grad aux in pynative mode
...
Merge pull request !40686 from chenzhuo/master_aux
2022-08-24 01:05:01 +00:00
chenzhuo
6775130f47
fix aux of grad in pynative mode
2022-08-23 14:32:28 +08:00
i-robot
ba031c7277
!40032 Add linearize ms_function test case
...
Merge pull request !40032 from ZhengXuegui/test_linearize
2022-08-16 19:27:02 +00:00
chenzhuo
5f39ebd788
support grad and value_and_grad with has_aux
2022-08-12 16:43:03 +08:00
chenzhuo
1c231080d0
fix weight in grad twice graph cache mistake
2022-08-11 17:10:26 +08:00
ZhengXuegui
9020c5a410
add linearize test case
2022-08-10 11:23:42 +08:00
i-robot
80d2d47ad7
!39321 Fix the lost mutable attr for ms_function inputs
...
Merge pull request !39321 from YuJianfeng/master
2022-08-02 09:12:00 +00:00
yujianfeng
4923cab2db
Fix the lost mutable attr for ms_function inputs
2022-08-01 14:42:49 +08:00
i-robot
d63fec2f41
!38660 Add linearize to optimize the performance of jvp under same inputs and fn but differnt v
...
Merge pull request !38660 from ZhengXuegui/add_linearize
2022-08-01 04:07:51 +00:00
ZhengXuegui
ed0d4a0ce4
add linearize for compute jvp
2022-07-29 15:04:00 +08:00
chenzhuo
4be29b46b5
support operator Tan in Taylor differentiation
2022-07-05 11:47:58 +08:00
zhousiyi
712a63d81e
lift multiple funcgraph of the same level J at one call
2022-06-28 09:06:06 +00:00
i-robot
e5d53e4c02
!34942 [bug] fix jet and grad input check error in graph mode
...
Merge pull request !34942 from chenzhuo/master
2022-06-07 07:35:04 +00:00
chenzhuo
4f7c9ffc80
fix jet input check error in graph mode
2022-06-07 10:08:16 +08:00
yujianfeng
1b7feccc37
Add pass to convert unused tuple output parameter to make_tuple
2022-06-02 09:50:06 +08:00
i-robot
8cabd14dcd
!35182 [bug] fix taylor rule of log
...
Merge pull request !35182 from chenzhuo/taylor_log
2022-05-31 01:33:29 +00:00
chenzhuo
73038f9d02
fix taylor rule of log
2022-05-30 16:26:15 +08:00
yujianfeng
d855c830d2
Change class Variable to mutable function
2022-05-16 09:05:32 +08:00