[V1] Make v1 more testable (#9888)

Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>

vllm/model_executor/models/llava_onevision.py

@@ -19,7 +19,7 @@ from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, DummyData,
                          InputContext, token_inputs)
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.quantization import QuantizationConfig
-from vllm.model_executor.layers.sampler import Sampler, SamplerOutput
+from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.utils import (cached_get_tokenizer,
@@ -437,7 +437,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal,
         if hasattr(self.language_model, "sampler"):
             return self.language_model.sampler
 
-        return Sampler()
+        return get_sampler()
 
     def _validate_image_sizes(self, data: torch.Tensor) -> torch.Tensor:
         expected_dims = (2, )