yuximiao
|
e99c0a48e6
|
support start profiler in the minddle of training.
|
4 years ago |
i-robot
|
875f35d6d8
|
!26541 Fix file name and field type changes generated by HCCL in profiler.
Merge pull request !26541 from casgj/master_hccl
|
4 years ago |
casgj
|
b15c09db6d
|
Fix file name and field type changes generated by HCCL.
|
4 years ago |
casgj
|
ce3916700d
|
cleancode in profiler.
|
4 years ago |
i-robot
|
ce00ee1ad1
|
!25367 use acl api to control profiling
Merge pull request !25367 from yanghaitao/yht_condation_start_profiler
|
4 years ago |
yanghaitao1
|
c94aa6b872
|
use profiler acl api instead
|
4 years ago |
i-robot
|
617485fa0b
|
!25838 Operator time data get the average value and remove the first step
Merge pull request !25838 from zangqx/profiling_gpu_permission
|
4 years ago |
mohammad
|
5c8ab5f60c
|
add MD Profiler Save()
|
4 years ago |
i-robot
|
d89a39c661
|
!25851 MD Profiling Start and Stop Support
Merge pull request !25851 from cathwong/ckw_prof_startstop4
|
4 years ago |
yelihua
|
cfa8c7a0e8
|
update OWNERS
|
4 years ago |
Cathy Wong
|
cdf3126459
|
MD Profiler Start Stop Support
|
4 years ago |
gaojing
|
4a4299d9c3
|
fix the error msg not clear when profiler path have chinese.
|
4 years ago |
臧庆香
|
235f655325
|
Get the average and remove the first step
|
4 years ago |
i-robot
|
b7af2b0cf7
|
!25470 Fix geting profiling job id fail when there are more than one JOB dirs in profiler output path
Merge pull request !25470 from ougongchang/fix_job_id
|
4 years ago |
ougongchang
|
34f6b50b17
|
Fix get profiling job id fail when there are more than one JOB dirs in profiler output path
|
4 years ago |
casgj
|
5af0365f73
|
clean code for profiler.
|
4 years ago |
huangbingjian
|
d7e97dd74a
|
change EXCEPTION level to CRITICAL level
|
4 years ago |
casgj
|
383746e323
|
Fix the error that operation name is null in aicore files.
|
4 years ago |
casgj
|
bf3dac8461
|
Fix profiler calling analysis twice.
|
4 years ago |
gaojing
|
e4b5d77b8e
|
fix the wrong value of average flops.
|
4 years ago |
i-robot
|
df82354be7
|
!24366 fix the error message that profiler does not support pynative mode.
Merge pull request !24366 from casgj/master
|
4 years ago |
yanghaitao1
|
a30b940ad3
|
add commiter to OWNERS files
|
4 years ago |
casgj
|
92052be1b6
|
fix the error message that profiler does not support pynative mode.
|
4 years ago |
i-robot
|
20b7cb94a0
|
!24077 adjust timeline description to make it easy to understand
Merge pull request !24077 from zangqx/profiling_gpu_permission
|
4 years ago |
gaojing
|
b78a13dd78
|
fix the error message that profiler does not support pynative mode.
|
4 years ago |
臧庆香
|
66e3775493
|
timeline description
|
4 years ago |
buxue
|
5418a45752
|
support cpu profiling and code check for cpu ops
|
4 years ago |
casgj
|
7c9b45f373
|
fix the error that the memory-related files generated are missing in profiler.
|
4 years ago |
i-robot
|
bc521674cc
|
!23473 modify the master build alarm
Merge pull request !23473 from zangqx/profiling_gpu_permission
|
4 years ago |
i-robot
|
75e7a5ebdc
|
!23482 Deal with case that the timeline 8001 show all communication operators lacks part of the communication operator time in profiler.
Merge pull request !23482 from zangqx/master_gaojing2
|
4 years ago |
臧庆香
|
ec03a76c66
|
modify the master build alarm
|
4 years ago |
i-robot
|
dc40225638
|
!23400 remove profiler if compiled with -s on
Merge pull request !23400 from yanghaitao/yht_remove_profiler
|
4 years ago |
zangqx
|
09a0392540
|
Deal with the case that the timeline operators lacks part of the communication operator in profiler.
|
4 years ago |
yanghaitao1
|
177f3f75bf
|
remove profiler if compiled with -s on
|
4 years ago |
i-robot
|
d7388b40ab
|
!23295 add dump and profiling warning log when task is not sink
Merge pull request !23295 from baihuawei/fixlog
|
4 years ago |
baihuawei
|
e1e11b9a47
|
fix some bugs
|
4 years ago |
zyhStack
|
55170ef7d9
|
Fix the problem with the default on communication performance switch
|
4 years ago |
yanghaitao1
|
83302ad23c
|
remove profiling parameters in set_context function
|
4 years ago |
i-robot
|
5fa582fef3
|
!23159 adjust the process node to which the HostCpuOps belongs
Merge pull request !23159 from zangqx/profiling_gpu_permission
|
4 years ago |
臧庆香
|
c4731c3efa
|
Adjust the process node to which the operator belongs
|
4 years ago |
gaojing
|
4f41b868bb
|
Raise error information for the case that the Profiler output path was not exists
|
4 years ago |
i-robot
|
7d6ff9d098
|
!23009 Modify the judge of multi devices training logic
Merge pull request !23009 from 张毅辉/Judge_multi_devices_training_logic
|
4 years ago |
zhangyihui
|
3a5b1f83f3
|
Modify and judge multi card training logic
|
4 years ago |
i-robot
|
b1be8dfd31
|
!22851 MD Profiling: Update Connector Init to Remove any existing file
Merge pull request !22851 from cathwong/ckw_mon_seq_pipelines_fix
|
4 years ago |
i-robot
|
9886c07c1c
|
!22459 Transform device_id to rank_id for cpu_profiler
Merge pull request !22459 from 张毅辉/Device_id_to_rank_id
|
4 years ago |
zhangyihui
|
6a36171a0e
|
transform device_id to rank_id for cpu_profiler
|
4 years ago |
臧庆香
|
c28bc7ccba
|
parser multiple st_track_data error
|
4 years ago |
Cathy Wong
|
bc85c606b8
|
MD Profiling: Update Connector Init to Remove any existing file
to fix sequential pipeline scenario.
|
4 years ago |
i-robot
|
5f9e9d96ec
|
!22542 Bugfix for scope-level flops data cannot display
Merge pull request !22542 from gzhcv/CommunicationOpNotOverlapped
|
4 years ago |
gzhcv
|
edb1b4798e
|
Bugfix for scope-level flops data cannot display
|
4 years ago |