From 74af2bbd901d82c6bc2583515c4388722d451f07 Mon Sep 17 00:00:00 2001 From: Alexander Matveev <59768536+alexm-neuralmagic@users.noreply.github.com> Date: Fri, 9 Aug 2024 12:35:49 -0400 Subject: [PATCH] [Bugfix] Fix reinit procedure in ModelInputForGPUBuilder (#7360) --- vllm/worker/model_runner.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/vllm/worker/model_runner.py b/vllm/worker/model_runner.py index 2731bddb..c71e2938 100644 --- a/vllm/worker/model_runner.py +++ b/vllm/worker/model_runner.py @@ -348,6 +348,8 @@ class ModelInputForGPUBuilder(ModelRunnerInputBuilderBase[ModelInputForGPU]): self.multi_modal_inputs = multi_modal_inputs self.prefix_cache_hit = prefix_cache_hit + self.n_seqs = len(self.seq_ids) + if not reinit: self.__post_init__()