[Bugfix] Fix typo in Pallas backend (vllm-project#5558)

Author: WoosukKwon

vllm/attention/backends/pallas.py

@@ -110,7 +110,7 @@ def __init__(
             raise NotImplementedError("TPU version must be 4 or higher.")
 
         self.megacore_mode = None
-        tpu_type = torch_xla.tpu.get_tp_groupu_env()["TYPE"].lower()
+        tpu_type = torch_xla.tpu.get_tpu_env()["TYPE"].lower()
         if not tpu_type.endswith("lite"):
             if self.num_kv_heads % 2 == 0:
                 self.megacore_mode = "kv_head"