Parse seed for vLLM (#602)

When seed is set through model_args, for example:

```bash
"pretrained=deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B,seed=123,dtype=bfloat16,max_model_length=38768,gpu_memory_utilization=0.8,tensor_parallel_size=1"
``

it is parsed as a string attribute in `VLLMModelConfig`, leading to an error during LLM initialization:

```python
model = LLM(**self.model_args)
```

This PR ensures that seed is correctly cast to an integer before passing it to the model, preventing initialization errors.

Co-authored-by: Nathan Habib <[email protected]>

Author: eldarkurtic

src/lighteval/models/vllm/vllm_model.py

@@ -182,7 +182,7 @@ def _create_auto_model(self, config: VLLMModelConfig, env_config: EnvConfig) ->
             "pipeline_parallel_size": int(config.pipeline_parallel_size),
             "max_model_len": self._max_length,
             "swap_space": 4,
-            "seed": config.seed,
+            "seed": int(config.seed),
             "max_num_seqs": int(config.max_num_seqs),
             "max_num_batched_tokens": int(config.max_num_batched_tokens),
         }