# tensorplay.distributed.elastic.utils.data API

Source: https://www.tensorplay.cn/docs/api/tensorplay.distributed.elastic.utils.data.html

## Classes 2

[#](#api-tensorplay.distributed.elastic.utils.data.CyclingIterator)

### CyclingIterator

class[Full reference ↗](/docs/generated/tensorplay.distributed.elastic.utils.data.CyclingIterator.html)

```python
class tensorplay.distributed.elastic.utils.data.CyclingIterator(*args, start_epoch: int = 0, n: int | None = None, generator_fn=None)
```

Wrap a finite iterable into an endless one.

Each cycle increments epoch; the underlying iterable is re-materialized
through its factory (or by calling iter on it again when it supports
that) so per-epoch shuffling keeps working.

```python
property epoch: int
```

Epoch number of the data currently being produced.

[#](#api-tensorplay.distributed.elastic.utils.data.ElasticDistributedSampler)

### ElasticDistributedSampler

class[Full reference ↗](/docs/generated/tensorplay.distributed.elastic.utils.data.ElasticDistributedSampler.html)

```python
class tensorplay.distributed.elastic.utils.data.ElasticDistributedSampler(dataset, *, start_rank: int = 0, start_epoch: int = 0)
```

Sampler for elastic jobs where num_replicas may change per epoch.

Unlike a fixed-shard distributed sampler, this one derives the covered
range from the epoch number so restarting with a different world size
never skips or repeats data beyond the wrap-around tail.