diff --git a/vllm/worker/tpu_worker.py b/vllm/worker/tpu_worker.py
index 71b4b38fb9d62..bbcc4d59ae1ca 100644
--- a/vllm/worker/tpu_worker.py
+++ b/vllm/worker/tpu_worker.py
@@ -163,8 +163,8 @@ class TPUWorker(LoRANotSupportedWorkerBase, LocalOrDistributedWorkerBase):
         usable_memory_size = int(total_memory_size *
                                  self.cache_config.gpu_memory_utilization)
         tpu_kv_cache_bytes = max(usable_memory_size - profiled, 0)
-        dtype_btyes = get_dtype_size(self.cache_dtype)
-        block_size_bytes = (dtype_btyes * self.cache_config.block_size *
+        dtype_bytes = get_dtype_size(self.cache_dtype)
+        block_size_bytes = (dtype_bytes * self.cache_config.block_size *
                             num_layers * 2 * head_size * num_kv_heads)
         num_tpu_blocks = tpu_kv_cache_bytes // block_size_bytes
         num_tpu_blocks = (num_tpu_blocks // 8) * 8  # Round down to 8.