Skip to content
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

Fix: Sparse tensors not updating #1914

Merged
merged 9 commits into from
May 23, 2022
Merged

Fix: Sparse tensors not updating #1914

Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
89bf66a
Fix do not updated sparse grads
Dipet Mar 9, 2022
7f54365
Merge branch 'master' into master
tjruwase Apr 26, 2022
f362660
Merge branch 'master' into master
jeffra Apr 26, 2022
216c8ce
Merge branch 'master' into master
jeffra Apr 27, 2022
b04c20a
Merge branch 'master' into master
Dipet May 12, 2022
f3acdfe
Merge branch 'master' into master
Dipet May 18, 2022
30ce968
Merge branch 'master' into master
tjruwase May 19, 2022
a1e2a96
Remove call .data for sparse grads
Dipet May 23, 2022
d6c34b2
Merge branch 'master' into master
Dipet May 23, 2022
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
3 changes: 2 additions & 1 deletion deepspeed/runtime/engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2169,7 +2169,8 @@ def _get_gradients_for_reduction(self):

grad_data = param.grad.data
if param_name in self.sparse_tensor_module_names or grad_data.is_sparse:
grad_data = SparseTensor(grad_data)
# Call param.grad without data to avoid problem with setting of updated grads
grad_data = SparseTensor(param.grad)

if is_moe_param(param):
expert_grads[param.group_name].append(grad_data)
Expand Down