mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
9e56378ef2
This populates the website with some basic documentation. It's far from ideal as we should include some basic usage example. Pull Request resolved: https://github.com/pytorch/pytorch/pull/92813 Approved by: https://github.com/wz337
70 lines
2.0 KiB
ReStructuredText
70 lines
2.0 KiB
ReStructuredText
.. role:: hidden
|
|
:class: hidden-section
|
|
|
|
Distributed Checkpoint - torch.distributed.checkpoint
|
|
=====================================================
|
|
|
|
|
|
Distributed Checkpoint (DCP) support loading and saving models from multiple ranks in parallel.
|
|
It handles load-time resharding which enables saving in one cluster topology and loading into another.
|
|
|
|
DCP is different than `torch.save` and `torch.load` in a few significant ways:
|
|
|
|
* It produces multiple files per checkpoint, with at least one per rank.
|
|
* It operates in place, meaning that the model should allocate its data first and DCP uses that storage instead.
|
|
|
|
The entrypoints to load and save a checkpoint are the following:
|
|
|
|
|
|
.. automodule:: torch.distributed.checkpoint
|
|
|
|
.. currentmodule:: torch.distributed.checkpoint
|
|
|
|
.. autofunction:: load_state_dict
|
|
.. autofunction:: save_state_dict
|
|
|
|
The following types define the IO interface used during checkpoint:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.StorageReader
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.StorageWriter
|
|
:members:
|
|
|
|
The following types define the planner interface used during checkpoint:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.LoadPlanner
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.LoadPlan
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.ReadItem
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.SavePlanner
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.SavePlan
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.WriteItem
|
|
:members:
|
|
|
|
We provide a filesystem based storage layer:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.FileSystemReader
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.FileSystemWriter
|
|
:members:
|
|
|
|
We provide default implementations of `LoadPlanner` and `SavePlanner` that
|
|
can handle all of torch.distributed constructs such as FSDP, DDP, ShardedTensor and DistributedTensor.
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.DefaultSavePlanner
|
|
:members:
|
|
|
|
.. autoclass:: torch.distributed.checkpoint.DefaultLoadPlanner
|
|
:members:
|