FIxing the AMD test failures caused by PR#16457 (vllm-project#17511)

Signed-off-by: Alexei V. Ivanov <[email protected]>
Signed-off-by: Mu Huai <[email protected]>

Author: Alexei-V-Ivanov-AMD

docker/Dockerfile.rocm

@@ -118,4 +118,3 @@ ENV TOKENIZERS_PARALLELISM=false
 ENV HIP_FORCE_DEV_KERNARG=1
 
 CMD ["/bin/bash"]
-

vllm/model_executor/layers/rotary_embedding.py

@@ -80,7 +80,7 @@ def _apply_rotary_emb(x: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor,
         is_neox_style: Whether to use the Neox-style or GPT-J-style rotary
             positional embeddings.
     """
-    if current_platform.is_cuda_alike():
+    if current_platform.is_cuda():
         return apply_rotary_emb(x.unsqueeze(0), cos, sin,
                                 not is_neox_style).squeeze(0)
     else: