[model][refactor] remove cuda hard code in models and layers (#13658)

vllm/model_executor/models/arctic.py

@@ -30,6 +30,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.model_executor.utils import set_weight_attrs
+from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.configs.arctic import ArcticConfig
 
@@ -138,13 +139,13 @@ class ArcticMoE(nn.Module):
                     torch.empty(self.num_experts,
                                 2 * self.intermediate_size,
                                 self.hidden_size,
-                                device="cuda",
+                                device=current_platform.device_type,
                                 dtype=self.params_dtype))
                 self.w2s = nn.Parameter(
                     torch.empty(self.num_experts,
                                 self.hidden_size,
                                 self.intermediate_size,
-                                device="cuda",
+                                device=current_platform.device_type,
                                 dtype=self.params_dtype))
             set_weight_attrs(self.ws, {
                 "weight_loader": self.weight_loader,