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

Jump to bottom

[UC] Fix compatible with npu #8409

Merged
merged 1 commit into from
May 9, 2024
Merged

[UC] Fix compatible with npu #8409

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
13 changes: 8 additions & 5 deletions paddlenlp/trainer/plugins/unified_checkpoint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,6 +93,11 @@
async_save_queue = []


DEST_PLACE = paddle.CPUPlace()
if paddle.device.is_compiled_with_cuda():
DEST_PLACE = paddle.CUDAPinnedPlace()


class UnifiedCheckpointOption(ExplicitEnum):
"""
"- skip_save_model_weight: do not save model weights when the masters weight exist\n"
Expand Down Expand Up @@ -1746,7 +1751,7 @@ def merge_tensor_parallel_with_shard(state_dict, tp_actions, all_filter_keys):
action = tp_actions.pop(key)
tensor = action(ret) if is_dst else None
else:
tensor = tensor._copy_to(paddle.CUDAPinnedPlace(), False) if is_dst else None
tensor = tensor._copy_to(DEST_PLACE, False) if is_dst else None

if is_dst:
state_dict_to_save[key] = tensor
Expand Down Expand Up @@ -1777,15 +1782,13 @@ def merge_tensor_parallel_for_optimizer(state_dict, tp_actions, all_filter_keys)
if model_key in tp_actions:
# for example: beta1, beta2
if tensor.numel().item() == 1:
tensor = (
tensor._copy_to(paddle.CUDAPinnedPlace(), False) if is_dst else None
) # Need broadcast when loaded
tensor = tensor._copy_to(DEST_PLACE, False) if is_dst else None # Need broadcast when loaded
else:
ret = distributed_gather(tensor, dst=j, group=tp_group, offload=False)
action = tp_actions[model_key]
tensor = action(ret) if is_dst else None
else:
tensor = tensor._copy_to(paddle.CUDAPinnedPlace(), False) if is_dst else None
tensor = tensor._copy_to(DEST_PLACE, False) if is_dst else None

if is_dst:
state_dict_to_save[filter_keys[i]] = tensor
Expand Down
Loading