megatron-fsdp 0.2.0.dev101882__tar.gz → 0.2.0.dev102308__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (17) hide show
  1. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/PKG-INFO +1 -1
  2. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/megatron_fsdp.py +4 -3
  3. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/package_info.py +1 -1
  4. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp.egg-info/PKG-INFO +1 -1
  5. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/README.md +0 -0
  6. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/__init__.py +0 -0
  7. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/distributed_data_parallel_config.py +0 -0
  8. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/fully_shard.py +0 -0
  9. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/param_and_grad_buffer.py +0 -0
  10. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/uneven_dtensor.py +0 -0
  11. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/utils.py +0 -0
  12. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp.egg-info/SOURCES.txt +0 -0
  13. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp.egg-info/dependency_links.txt +0 -0
  14. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp.egg-info/requires.txt +0 -0
  15. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp.egg-info/top_level.txt +0 -0
  16. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/pyproject.toml +0 -0
  17. {megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/setup.cfg +0 -0
{megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: megatron-fsdp
3
- Version: 0.2.0.dev101882
3
+ Version: 0.2.0.dev102308
4
4
  Summary: **Megatron-FSDP** is an NVIDIA-developed PyTorch extension that provides a high-performance implementation of Fully Sharded Data Parallelism (FSDP)
5
5
  Author-email: NVIDIA <nemo-toolkit@nvidia.com>
6
6
  Maintainer-email: NVIDIA <nemo-toolkit@nvidia.com>
{megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/megatron_fsdp.py RENAMED
@@ -898,9 +898,10 @@ class MegatronFSDP(torch.nn.Module):
898
898
 
899
899
  # Register pre state_dict hook to ensure that the module parameters are
900
900
  # distributed before saving the state_dict.
901
- self._state_dict_pre_hook = self.module.register_state_dict_pre_hook(
902
- lambda *args, **kwargs: self._replace_param_with_distributed_if_needed()
903
- )
901
+ for name, module in self.named_modules():
902
+ module.register_state_dict_pre_hook(
903
+ lambda *args, **kwargs: self._replace_param_with_distributed_if_needed()
904
+ )
904
905
 
905
906
  @contextmanager
906
907
  def no_sync(self):
{megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp/package_info.py RENAMED
@@ -4,7 +4,7 @@
4
4
  MAJOR = 0
5
5
  MINOR = 2
6
6
  PATCH = 0
7
- PRE_RELEASE = '0.dev101882'
7
+ PRE_RELEASE = '0.dev102308'
8
8
 
9
9
  # Use the following formatting: (major, minor, patch, pre-release)
10
10
  VERSION = (MAJOR, MINOR, PATCH, PRE_RELEASE)
{megatron_fsdp-0.2.0.dev101882 → megatron_fsdp-0.2.0.dev102308}/megatron_fsdp.egg-info/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: megatron-fsdp
3
- Version: 0.2.0.dev101882
3
+ Version: 0.2.0.dev102308
4
4
  Summary: **Megatron-FSDP** is an NVIDIA-developed PyTorch extension that provides a high-performance implementation of Fully Sharded Data Parallelism (FSDP)
5
5
  Author-email: NVIDIA <nemo-toolkit@nvidia.com>
6
6
  Maintainer-email: NVIDIA <nemo-toolkit@nvidia.com>