Skip to content

ZeRO-2 CPU_offload #4

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

Merged
jren73 merged 1 commit into from
Aug 11, 2020
Merged

ZeRO-2 CPU_offload #4

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
29 changes: 19 additions & 10 deletions deepspeed/pt/deepspeed_zero_optimizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@
import math
from torch._six import inf
from torch.autograd import Variable
import collections

from deepspeed.pt.loss_scaler import LossScaler, DynamicLossScaler
from deepspeed.pt.deepspeed_utils import see_memory_usage, is_model_parallel_parameter
Expand Down Expand Up @@ -120,11 +121,18 @@ def async_migrate_to(obj, dev, main_stream=None):
return obj


def async_copy_to(v_in, v_out, main_stream=None):
#in_dev = in.device()
#out_dev = out.device()
print(v_in.size())
print(v_out.size())
def async_copy_to(obj, dev, main_stream=None):
if torch.is_tensor(obj):
obj = Variable(obj)
if isinstance(obj, Variable):
target = torch.empty_like(obj, device=dev).copy_(obj)
if main_stream is not None:
target.data.record_stream(main_stream)
return target
elif isinstance(obj, collections.Mapping):
return {k: async_copy_to(o, dev, main_stream) for k, o in obj.items()}
elif isinstance(obj, collections.Sequence):
return [async_copy_to(o, dev, main_stream) for o in obj]



Expand Down Expand Up @@ -1257,16 +1265,15 @@ def step_with_cpuoffload(self, closure=None):
#jie: transform the grad back to CPU
start, end = self.get_partition_fp32_group_index(i)
with torch.cuda.stream(self.migration_stream):
#self.single_partition_of_fp32_groups[start].grad = single_grad_partition.detach().clone()
self.single_partition_of_fp32_groups[start:end].grad = torch.tensor(single_grad_partition.detach(),
dtype=self.single_partition_of_fp32_groups[i].dtype)
self.single_partition_of_fp32_groups[start:end].grad = async_copy_to(single_grad_partition,'cpu', self.migration_stream)
#self.single_partition_of_fp32_groups[start:end].grad = torch.tensor(single_grad_partition.detach(),
# dtype=self.single_partition_of_fp32_groups[i].dtype)
#release all the gradient since we have already created a necessary copy in dp_grad_partition
self.free_grad_in_param_list(self.params_in_partition[i])

self.averaged_gradients[i] = None

#single_partition_grad_groups.append(single_grad_partition)

#jie: is syncization necessary???
self.cpu_computation_stream.wait_stream(self.migration_stream)
timers('optimizer_step').start()
with torch.cuda.stream(self.cpu_computation_stream):
Expand All @@ -1286,6 +1293,8 @@ def step_with_cpuoffload(self, closure=None):
for i, fp16_partitions in enumerate(self.parallel_partitioned_fp16_groups):
start, end = self.get_partition_fp32_group_index(i)
fp16_partitions[partition_id].data.copy_(self.single_partition_of_fp32_groups[start:end].data)
return

timers('optimizer_step').stop()

timers('optimizer_allgather').start()
Expand Down