diff --git a/.buildkite/vllm_lkg.version b/.buildkite/vllm_lkg.version
index 86b654098b..93bc509737 100644
--- a/.buildkite/vllm_lkg.version
+++ b/.buildkite/vllm_lkg.version
@@ -1 +1 @@
-8c0b6267d7fa5c8a07e318809180fc021a0afbf2
+d9408ffba3c8da5e289b5695d507707afce10a2f
diff --git a/tests/runner/test_input_batch.py b/tests/runner/test_input_batch.py
index 4a218cdb9c..8fe0cbdf0b 100644
--- a/tests/runner/test_input_batch.py
+++ b/tests/runner/test_input_batch.py
@@ -261,6 +261,7 @@ def meta_eq(a: PoolingMetadata, b: PoolingMetadata):
         PoolingMetadata(
             prompt_lens=torch.tensor([], dtype=torch.int32),
             prompt_token_ids=None,
+            prompt_token_ids_cpu=None,
             pooling_params=[],
             pooling_states=[],
         ),
@@ -281,6 +282,7 @@ def meta_eq(a: PoolingMetadata, b: PoolingMetadata):
         PoolingMetadata(
             prompt_lens=torch.tensor([10], dtype=torch.int32),
             prompt_token_ids=None,
+            prompt_token_ids_cpu=None,
             pooling_params=[pooling_param],
             pooling_states=[pooling_state],
         ),
@@ -292,6 +294,7 @@ def meta_eq(a: PoolingMetadata, b: PoolingMetadata):
         PoolingMetadata(
             prompt_lens=torch.tensor([], dtype=torch.int32),
             prompt_token_ids=None,
+            prompt_token_ids_cpu=None,
             pooling_params=[],
             pooling_states=[],
         ),
diff --git a/tpu_inference/executors/ray_distributed_executor.py b/tpu_inference/executors/ray_distributed_executor.py
index 3042bf90b6..4dbea3495e 100644
--- a/tpu_inference/executors/ray_distributed_executor.py
+++ b/tpu_inference/executors/ray_distributed_executor.py
@@ -32,6 +32,7 @@
 from vllm.v1.executor.ray_distributed_executor import \
     RayDistributedExecutor as RayDistributedExecutorV1
 from vllm.v1.executor.ray_executor import RayWorkerMetaData
+from vllm.v1.executor.ray_utils import WORKER_SPECIFIC_ENV_VARS
 from vllm.v1.executor.ray_utils import RayWorkerWrapper as RayWorkerWrapperV1
 from vllm.v1.executor.ray_utils import _wait_until_pg_ready
 from vllm.v1.outputs import ModelRunnerOutput
@@ -363,7 +364,7 @@ def sort_by_driver_then_worker_ip(item: RayWorkerMetaData):
 
         # Environment variables to copy from driver to workers
         env_vars_to_copy = get_env_vars_to_copy(
-            exclude_vars=self.WORKER_SPECIFIC_ENV_VARS,
+            exclude_vars=WORKER_SPECIFIC_ENV_VARS,
             additional_vars=set(current_platform.additional_env_vars),
             destination="workers")
 
@@ -483,7 +484,7 @@ class RayWorkerWrapper(RayWorkerWrapperV1):
     Ray worker wrapper for TPU.
 
     The implementation is similar to vllm/v1/executor/ray_utils.py
-    
+
     _is_intermediate_tensors: check whether the output is JaxIntermediateTensors.
     _is_last_rank: check whether this Ray worker is the last PP stage.
     """
diff --git a/tpu_inference/runner/input_batch.py b/tpu_inference/runner/input_batch.py
index 3e658c0d2b..392328bcb9 100644
--- a/tpu_inference/runner/input_batch.py
+++ b/tpu_inference/runner/input_batch.py
@@ -163,6 +163,7 @@ def get_pooling_metadata(self) -> PoolingMetadata:
             prompt_lens=torch.from_numpy(
                 self.num_prompt_tokens[:self.num_reqs]),
             prompt_token_ids=None,
+            prompt_token_ids_cpu=None,
             pooling_params=pooling_params,
             pooling_states=pooling_states,
         )