> 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-generation-vllm-config/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.generation.vllm.config

## Contenu du Module

### Classes

| Nom                                                                     | Description                                                                                                                                                                                                                                                                                                                                                                                                |
| ----------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [`VllmSpecificArgs`](#nemorlmodelsgenerationvllmconfigvllmspecificargs) | dict() -> nouveau dictionnaire vide dict(mapping) -> nouveau dictionnaire initialisé à partir des paires (clé, valeur) d'un objet de mappage dict(iterable) -> nouveau dictionnaire initialisé comme suit : d =  for k, v in iterable: d\[k] = v dict(\*\*kwargs) -> nouveau dictionnaire initialisé avec les paires nom=valeur dans la liste des arguments de mots-clés. Par exemple : dict(one=1, two=2) |
| [`VllmConfig`](#nemorlmodelsgenerationvllmconfigvllmconfig)             | Configuration pour la génération.                                                                                                                                                                                                                                                                                                                                                                          |

### API

```python
class nemo_rl.models.generation.vllm.config.VllmSpecificArgs
```

**Bases**: `typing.TypedDict`

```python
tensor_parallel_size: int
```

**Valeur**: `None`

```python
pipeline_parallel_size: int
```

**Valeur**: `None`

```python
enable_expert_parallel: bool
```

**Valeur**: `None`

```python
gpu_memory_utilization: float
```

**Valeur**: `None`

```python
max_model_len: int
```

**Valeur**: `None`

```python
skip_tokenizer_init: bool
```

**Valeur**: `None`

```python
async_engine: bool
```

**Valeur**: `None`

```python
load_format: typing.NotRequired[str]
```

**Valeur**: `None`

```python
precision: typing.NotRequired[str]
```

**Valeur**: `None`

```python
enforce_eager: typing.NotRequired[bool]
```

**Valeur**: `None`

```python
class nemo_rl.models.generation.vllm.config.VllmConfig
```

**Bases**: `nemo_rl.models.generation.interfaces.GenerationConfig`

```python
vllm_cfg: nemo_rl.models.generation.vllm.config.VllmSpecificArgs
```

**Valeur**: `None`

```python
vllm_kwargs: typing.NotRequired[dict[str, typing.Any]]
```

**Valeur**: `None`