Remove inference_mode() from platforms.hpu (#1691)

Inference_mode() is causing recompilations with t.compile - we don't
need it as we already put inference_mode on particular functions in
model runner. It was introduced by Rebase 0.9.0.1
(#1507) - previously we didn't
have such call.

Author: jkaniecki

vllm/platforms/hpu.py

@@ -68,10 +68,6 @@ def is_async_output_supported(cls, enforce_eager: Optional[bool]) -> bool:
     def get_device_name(cls, device_id: int = 0) -> str:
         return cls.device_name
 
-    @classmethod
-    def inference_mode(cls):
-        return torch.no_grad()
-
     @classmethod
     def check_and_update_config(cls, vllm_config: VllmConfig) -> None: