> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://fr.nvidia-localization.ferndocs.com/fr/nemo/rl/latest/nemo-rl-models-policy-dtensor-policy-worker-v2/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://fr.nvidia-localization.ferndocs.com/_mcp/server.

# nemo\_rl.models.policy.dtensor\_policy\_worker\_v2

## Contenu du Module

### Classes

| Nom                                                                                      | Description |
| ---------------------------------------------------------------------------------------- | ----------- |
| [`DTensorPolicyWorkerV2`](#nemorlmodelspolicydtensorpolicyworkerv2dtensorpolicyworkerv2) | Aucun       |

### API

```python
class nemo_rl.models.policy.dtensor_policy_worker_v2.DTensorPolicyWorkerV2(config: nemo_rl.models.policy.PolicyConfig, tokenizer: transformers.AutoTokenizer, processor: typing.Optional[transformers.AutoProcessor] = None, weights_path: typing.Optional[str] = None, optimizer_path: typing.Optional[str] = None, init_optimizer: bool = True, init_reference_model: bool = True, **kwargs: typing.Any)
```

```python
__repr__() -> str
```

Personnalise le préfixe de l'acteur dans les journaux Ray.

Cela facilite l'identification du worker produisant des messages de journal spécifiques.

```python
_apply_temperature_scaling(logits: torch.Tensor) -> torch.Tensor
```

```python
init_collective(
    ip: str,
    port: int,
    world_size: int
) -> None
```

Initialiser la communication collective.

```python
is_alive() -> bool
```

```python
reset_peak_memory_stats() -> None
```

```python
get_gpu_info() -> dict[str, typing.Any]
```

Retourne des informations sur le GPU utilisé par ce worker.

\[... rest of the document continues with the same translation pattern ...]