github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-21 09:29:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

[async io]supoort async io (#6137)

Browse Source 
* support async optimizer save/load

* fix

* fix

* support pin mem

* Update low_level_zero_plugin.py

* fix

* fix

* fix

* fix

* fix
This commit is contained in:
flybird11111
2024-11-18 17:52:24 +08:00
committed by Hongxin Liu
parent b90835bd32
commit eb69e640e5
15 changed files with 374 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
colossalai/booster/booster.py
View File

@@ -359,6 +359,7 @@ class Booster:
gather_dtensor: bool = True,
prefix: Optional[str] = None,
size_per_shard: int = 1024,
use_async: bool = False,
) -> None:
"""
Save optimizer to checkpoint.
@@ -374,7 +375,9 @@ class Booster:
names to compose the keys in state_dict. Defaults to None.
size_per_shard (int, optional): Maximum size of checkpoint shard file in MB. This is useful only when ``shard=True``. Defaults to 1024.
"""
self.checkpoint_io.save_optimizer(optimizer, checkpoint, shard, gather_dtensor, prefix, size_per_shard)
self.checkpoint_io.save_optimizer(
optimizer, checkpoint, shard, gather_dtensor, prefix, size_per_shard, use_async=use_async
)
def save_lr_scheduler(self, lr_scheduler: LRScheduler, checkpoint: str) -> None:
"""Save lr scheduler to checkpoint.