From 430783018cbfad69c6ff3a52479bf7b556b65247 Mon Sep 17 00:00:00 2001 From: Siyuan Liu Date: Sun, 11 May 2025 21:58:33 -0700 Subject: [PATCH] [Bugfix][TPU] Use np array when updating cache slot_mapping (#17971) Signed-off-by: Siyuan Liu --- vllm/v1/worker/tpu_model_runner.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/v1/worker/tpu_model_runner.py b/vllm/v1/worker/tpu_model_runner.py index 983f8707a2455..687dabee22904 100644 --- a/vllm/v1/worker/tpu_model_runner.py +++ b/vllm/v1/worker/tpu_model_runner.py @@ -531,7 +531,7 @@ class TPUModelRunner(LoRAModelRunnerMixin): np.add(block_numbers * self.block_size, block_offsets, out=self.input_batch.block_table. - slot_mapping_cpu[:total_num_scheduled_tokens]) + slot_mapping_np[:total_num_scheduled_tokens]) # Prepare the attention metadata. self.query_start_loc_np[0] = 0