# tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook API

Source: https://www.tensorplay.cn/docs/api/tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.html

## Functions 1

[#](#api-tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.post_localSGD_hook)

### post_localSGD_hook

function[Full reference ↗](/docs/generated/tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.post_localSGD_hook.html)

```python
tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.post_localSGD_hook(state: PostLocalSGDState, bucket: GradBucket)
```

Run post-localSGD algorithm.

This DDP communication hook is used for running post-localSGD algorithm,
by combining with a model averaging component (e.g.,
[PeriodicModelAverager](/docs/generated/tensorplay.distributed.algorithms.model_averaging.PeriodicModelAverager.html#tensorplay.distributed.algorithms.model_averaging.PeriodicModelAverager))
that runs after the optimizer step.

Parameters:

- state ([PostLocalSGDState](/docs/generated/tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.PostLocalSGDState.html#tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.PostLocalSGDState)) – State information to run post-localSGD. Users mainly need to tune start_localSGD_iter to determine when to start local SGD.

- bucket (dist.GradBucket) – Bucket that stores a 1D flattened gradient tensor that batches multiple per-variable tensors. Note that since DDP comm hook only supports single process single device mode, only exactly one tensor is stored in this bucket.

Returns:

Future handler of the communication, which updates the gradients in place.

Example::

```
>>> # xdoctest: +SKIP
>>> state = PostLocalSGDState(process_group=process_group, subgroup=subgroup,
                          start_localSGD_iter=10)
>>> ddp_model.register_comm_hook(state, post_localSGD_hook)
>>> # Also need to establish a model averaging module and run model averaging after ``optimizer.step()``.
```

## Classes 1

[#](#api-tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.PostLocalSGDState)

### PostLocalSGDState

class[Full reference ↗](/docs/generated/tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.PostLocalSGDState.html)

```python
class tensorplay.distributed.algorithms.ddp_comm_hooks.post_localSGD_hook.PostLocalSGDState(process_group, subgroup, start_localSGD_iter, post_local_gradient_allreduce=True)
```

Store state for all-reducing gradients globally until given step, then locally after.

Stores the state for all-reducing gradients globally using process_group until step start_localSGD_iter,
and all-reducing gradients locally using subgroup afterwards.

If process_group is None, the global process group will be used.
If subgroup is None, the intra-node process group on each machine will be used.

Additionally, post_local_gradient_allreduce may be worth tuning,
because both true and false may give a faster convergence.

```python
maybe_increase_iter(bucket)
```

Track iterations and trigger log message at start of local SGD.