fix tpu torch compile error

Signed-off-by: Chenyaaang <chenyangli@google.com>

Author: Chenyaaang

vllm/config/vllm.py

@@ -330,9 +330,12 @@ def __post_init__(self):
             + self.compilation_config.custom_ops.count("all")
             == 0
         ):
+            from vllm.platforms import current_platform
+
             if (
                 self.compilation_config.level > 0
                 and self.compilation_config.backend != "eager"
+                and not current_platform.is_tpu()
             ):
                 self.compilation_config.custom_ops.append("none")
             else:

vllm/platforms/tpu.py

@@ -139,8 +139,9 @@ def check_and_update_config(cls, vllm_config: VllmConfig) -> None:
             )
             compilation_config.cudagraph_mode = CUDAGraphMode.NONE
 
-        if compilation_config.backend == "":
-            compilation_config.backend = "openxla"
+        # Note: the default backend is set to inductor now
+        # we want to overwrite to openxla to execute the ops properly on TPU.
+        compilation_config.backend = "openxla"
 
         assert vllm_config.speculative_config is None, (
             "TPU does not support speculative decoding"