From 0999b7224a0f7334cdc9db2fb7be725380978647 Mon Sep 17 00:00:00 2001 From: zhuhaoran Date: Sat, 6 Dec 2025 13:28:08 +0800 Subject: [PATCH] fix: apply suggestions from @njhill Signed-off-by: zhuhaoran --- vllm/v1/worker/gpu_input_batch.py | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) diff --git a/vllm/v1/worker/gpu_input_batch.py b/vllm/v1/worker/gpu_input_batch.py index 0f523eacfa4bf..ff71c6d54179d 100644 --- a/vllm/v1/worker/gpu_input_batch.py +++ b/vllm/v1/worker/gpu_input_batch.py @@ -950,10 +950,9 @@ class InputBatch: assert self.async_copy_ready_event is not None self.async_copy_ready_event.synchronize() sampled_token_ids = self.sampled_token_ids_cpu.tolist() - # Replace placeholder token id with actual sampled id. - req_output_token_ids[-len(sampled_token_ids[prev_index]) :] = ( - sampled_token_ids[prev_index] - ) + # Replace placeholder token id(s) with actual sampled id(s). + if sampled_ids := sampled_token_ids[prev_index]: + req_output_token_ids[-len(sampled_ids) :] = sampled_ids @property def num_reqs(self) -> int: