Skip to content

fmoe with deepspeed #162

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
KimmiShi wants to merge 3 commits into
base: master
Choose a base branch
from
Open

fmoe with deepspeed #162

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
0edecc5
support nccl path from env
KimmiShi Jun 27, 2023
4d29af3
bf16
KimmiShi Jun 28, 2023
194cda9
add kwargs for gate
KimmiShi Jun 29, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion cuda/fastermoe/smart_schedule.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,7 +155,7 @@ torch::Tensor _smart_sch_backward(
auto global_grad_in = grad_out.new_zeros({global_batch_size, d_model});
auto grad_in = grad_out.new_zeros({buf_batch_size, d_model});

AT_DISPATCH_FLOATING_TYPES_AND_HALF(grad_out.scalar_type(),
AT_DISPATCH_FLOATING_TYPES_AND2(at::ScalarType::Half, at::ScalarType::BFloat16,grad_out.scalar_type(),
"fmoe_cuda_smartsch_backward", ([&] {
fmoe_cuda_fused_backward_impl(
backward_fn,
Expand Down
6 changes: 3 additions & 3 deletions fmoe/gates/gshard_gate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,9 +11,9 @@

class GShardGate(NaiveGate):
def __init__(self, d_model, num_expert, world_size,
topk=2, capacity=(1.2, 2.4), random_routing=True):
assert topk == 2, 'topk should be 2 in gshard'
super().__init__(d_model, num_expert, world_size, top_k=2)
topk=2, capacity=(1, 1), random_routing=False):
# assert topk == 2, 'topk should be 2 in gshard'
super().__init__(d_model, num_expert, world_size, top_k=topk)
self.capacity = capacity
self.random_routing = random_routing

Expand Down
3 changes: 2 additions & 1 deletion fmoe/layers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,6 +117,7 @@ def __init__(
gate_hook=None,
mask=None,
mask_dict=None,
gate_kwargs={}
):
super().__init__()
self.num_expert = num_expert
Expand Down Expand Up @@ -145,7 +146,7 @@ def __init__(
else:
self.experts_fused = True

self.gate = gate(d_model, num_expert, world_size, top_k)
self.gate = gate(d_model, num_expert, world_size, top_k, **gate_kwargs)
self.gate_hook = gate_hook
self.mask = mask
self.mask_dict = mask_dict
Expand Down
20 changes: 13 additions & 7 deletions setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,11 +7,11 @@
ext_libs = []

authors = [
'Jiaao He',
'Jiezhong Qiu',
'Aohan Zeng',
'Tiago Antunes',
'Jinjun Peng',
'Jiaao He',
'Jiezhong Qiu',
'Aohan Zeng',
'Tiago Antunes',
'Jinjun Peng',
'Qin Li',
'Mingshu Zhai'
]
Expand All @@ -37,6 +37,11 @@
else:
define_macros=[]

include_dirs = []
if os.environ.get("NCCL_PATH"):
include_dirs.append(os.environ.get("NCCL_PATH")+'/include')
nccl_lib_path = os.environ.get("NCCL_PATH")+'/lib'
os.environ['LIBRARY_PATH'] = nccl_lib_path+':'+os.environ.get('LIBRARY_PATH','')

if __name__ == '__main__':
setuptools.setup(
Expand All @@ -50,7 +55,7 @@
packages=['fmoe', 'fmoe.megatron', 'fmoe.gates', 'fmoe.fastermoe'],
ext_modules=[
CUDAExtension(
name='fmoe_cuda',
name='fmoe_cuda',
sources=[
'cuda/stream_manager.cpp',
'cuda/local_exchange.cu',
Expand All @@ -65,7 +70,8 @@
'cxx': cxx_flags,
'nvcc': cxx_flags
},
libraries=ext_libs
libraries=ext_libs,
include_dirs=include_dirs
)
],
cmdclass={
Expand Down