Skip to content

[shardformer] bert support sequence parallel. #4455

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
flybird11111 merged 4 commits into from
Aug 18, 2023
Merged

[shardformer] bert support sequence parallel. #4455

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4c5320f
[shardformer] bert support sequence parallel
flybird11111 Aug 16, 2023
094b58d
[shardformer] bert support sequence parallel
flybird11111 Aug 18, 2023
9035586
Merge branch 'feature/shardformer' into update-seq-bert
flybird11111 Aug 18, 2023
6bd6b99
[shardformer] bert support sequence parallel
flybird11111 Aug 18, 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
6 changes: 1 addition & 5 deletions colossalai/shardformer/layer/_operation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -154,7 +154,7 @@ class _LinearWithGatherForwardReduceScatterBackward(torch.autograd.Function):
"""

@staticmethod
def forward(ctx, input_, weight, bias, process_group, async_grad_reduce_scatter, dim, overlap):
def forward(ctx, input_, weight, bias, process_group, async_grad_reduce_scatter, dim, overlap=True):
ctx.save_for_backward(input_, weight)
ctx.use_bias = bias is not None
ctx.process_group = process_group
Expand Down Expand Up @@ -217,9 +217,7 @@ def backward(ctx, grad_output):
# do all gather in default stream
input_ = input_.contiguous()
world_size = dist.get_world_size(process_group)
rank = dist.get_rank(process_group)
tensor_list = [torch.empty_like(input_) for _ in range(world_size)]
tensor_list[rank] = input_
gather_handle = dist.all_gather(tensor_list, input_, group=process_group, async_op=True)

# calculate gradient in calculate_stream
Expand Down Expand Up @@ -469,9 +467,7 @@ def _gather(input_, dim=-1, process_group=None):

# all gather
input_ = input_.contiguous()
rank = dist.get_rank(process_group)
tensor_list = [torch.empty_like(input_) for _ in range(world_size)]
tensor_list[rank] = input_
torch.distributed.all_gather(tensor_list, input_, group=process_group)

# concat
Expand Down
Loading