issues
search
microsoft
/
tutel
Tutel MoE: An Optimized Mixture-of-Experts Implementation
MIT License
723
stars
93
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
support example: tutel.examples.moe_cifar10
#145
ghostplant
closed
2 years ago
0
distinguish cpu/gpu switching
#144
ghostplant
closed
2 years ago
0
fix gradient update in MNIST example
#143
ghostplant
closed
2 years ago
0
add example: tutel.examples.moe_mnist
#142
ghostplant
closed
2 years ago
0
enable jit_compiler to support dynamic blocks
#141
ghostplant
closed
2 years ago
0
enable regression tests for main branch
#140
ghostplant
closed
2 years ago
0
support `activation_fn_with_self` in expert option
#139
ghostplant
closed
2 years ago
0
Tutel with pytorch automatic mixed precision package
#138
MiZhenxing
opened
2 years ago
2
more descriptions of capacity_factor
#137
ghostplant
closed
2 years ago
0
What is the purpose of the "use_2dh" option?
#136
ymjiang
closed
2 years ago
4
allow negative cap-factor
#135
ghostplant
closed
2 years ago
0
Fix Bug - Fix group can be None in global_expert_count
#134
yzygitzh
closed
2 years ago
1
try nvcc first when command exists
#133
ghostplant
closed
2 years ago
0
module 'tutel_custom_kernel' has no attribute 'inject_source'
#132
LisaWang0306
closed
2 years ago
10
add --eval option; add background mode on a2a;
#131
ghostplant
closed
2 years ago
0
update scratch examples
#130
ghostplant
closed
2 years ago
0
update UT expected step time
#129
EricWangCN
closed
2 years ago
0
allow unknown GPU to run "./setup.py test"
#128
ghostplant
closed
2 years ago
0
fix backward gating function
#127
ghostplant
closed
2 years ago
0
cancel computing grad on data input
#126
ghostplant
closed
2 years ago
0
add allreduce_degree option for profiling purpose
#125
ghostplant
closed
2 years ago
0
fix compatibility with ROCm 4.5
#124
ghostplant
closed
2 years ago
0
use simplified function name in net.py
#123
ghostplant
closed
2 years ago
0
use simplified function name in net.py
#122
ghostplant
closed
2 years ago
0
add helloworld_from_scratch.py
#121
ghostplant
closed
2 years ago
0
update README.md
#120
ghostplant
closed
2 years ago
0
declare seperate interfaces
#119
ghostplant
closed
2 years ago
0
add Unit Test of CPU kernel
#118
EricWangCN
closed
2 years ago
0
keep output dims fixed for fast_encode/fast_decode
#117
ghostplant
closed
2 years ago
0
Fix ROCm hipify issue for backward compatibility
#116
abuccts
closed
2 years ago
0
update README.md for cpu launch
#115
ghostplant
closed
2 years ago
0
update examples for cpu support
#114
ghostplant
closed
2 years ago
0
reimplement sparse cpu kernels
#113
ghostplant
closed
2 years ago
0
add up depend list into custom ops
#112
ghostplant
closed
2 years ago
0
update parted settings
#111
ghostplant
closed
2 years ago
0
Fix apply keyword restriction
#110
ghostplant
closed
2 years ago
0
add legacy partition tools
#109
ghostplant
closed
2 years ago
0
update communication libs
#108
ghostplant
closed
2 years ago
0
rename is_postnorm to is_postscore
#107
ghostplant
closed
2 years ago
0
rename is_postnorm to is_postscore
#106
ghostplant
closed
2 years ago
0
update data flow for auto parallel
#105
ghostplant
closed
2 years ago
0
hybrid model_parallel into data_parallel session
#104
ghostplant
closed
2 years ago
0
hybrid model_parallel into data_parallel session
#103
ghostplant
closed
2 years ago
0
Add performance figures
#102
EricWangCN
closed
2 years ago
0
Add performance figures
#101
EricWangCN
closed
2 years ago
0
Merge A2A FFN overlapping and 2DH A2A
#100
yzygitzh
closed
2 years ago
0
handle occupancy compat for rocm4.2
#99
ghostplant
closed
2 years ago
0
why Deepspeed MoE Top-2 Gate dosen't integrate Tutel acceleration
#98
Satan012
closed
2 years ago
1
simplify all different usages into top-k usage
#97
ghostplant
closed
2 years ago
0
Error:Exception: MoE JIT is designed to work on sample size = 800, while receiving sample size = 1600 (> 800)
#96
Satan012
opened
2 years ago
2
Previous
Next