| .. |
|
ascend
|
get rank id when set hccl env for single card train
|
2021-08-17 12:10:45 +08:00 |
|
cpu
|
!20995 pyfunc cpu kernel
|
2021-08-03 04:51:20 +00:00 |
|
executor
|
fix code review bugs
|
2021-07-28 11:28:57 +08:00 |
|
gpu
|
using kernel pool to share the compiling results when running on multi
|
2021-08-02 15:25:17 +08:00 |
|
CMakeLists.txt
|
ascend support mpi
|
2021-07-30 16:22:48 +08:00 |
|
bucket.cc
|
fix some pclint warnings in master
|
2021-04-22 17:06:16 +08:00 |
|
bucket.h
|
unify runtime for PyNative distributed mode
|
2021-05-20 11:39:14 +08:00 |
|
convert_tensor_utils.cc
|
unified runtime codedex fixed
|
2021-07-30 13:00:38 +08:00 |
|
convert_tensor_utils.h
|
unified runtime codedex fixed
|
2021-07-30 13:00:38 +08:00 |
|
device_address.h
|
fix bug of output none and share weight
|
2021-07-13 22:06:52 +08:00 |
|
kernel_adjust.cc
|
clean code
|
2021-07-27 10:30:21 +08:00 |
|
kernel_adjust.h
|
clean code
|
2021-07-19 21:08:52 +08:00 |
|
kernel_info.cc
|
clean code
|
2021-07-27 10:30:21 +08:00 |
|
kernel_info.h
|
unified runtime codedex fixed
|
2021-07-30 13:00:38 +08:00 |
|
kernel_runtime.cc
|
ascend support nontask sink
|
2021-08-03 20:55:43 +08:00 |
|
kernel_runtime.h
|
use op transdata to improve performance in print process
|
2021-07-06 15:09:18 +08:00 |
|
kernel_runtime_manager.cc
|
clean code
|
2021-07-27 10:30:21 +08:00 |
|
kernel_runtime_manager.h
|
adapt for multi-frontend cj
|
2021-06-29 01:14:52 +08:00 |
|
launch_kernel.cc
|
clean code
|
2021-07-27 10:30:21 +08:00 |
|
launch_kernel.h
|
add op atomic clean to clear input addr in launch allreduce
|
2021-03-25 11:19:23 +08:00 |
|
launch_mul.cc
|
clean code
|
2021-07-27 10:30:21 +08:00 |
|
launch_mul.h
|
add op atomic clean to clear input addr in launch allreduce
|
2021-03-25 11:19:23 +08:00 |
|
memory_manager.cc
|
change neighbor exchange to all to all
|
2021-08-01 18:21:28 +08:00 |
|
memory_manager.h
|
change neighbor exchange to all to all
|
2021-08-01 18:21:28 +08:00 |
425b1168ad