From f2dd24ed34a77fcb6e868e24d0a60cdcf6491554 Mon Sep 17 00:00:00 2001 From: Luo Cheng Date: Thu, 21 Mar 2024 06:23:52 +0100 Subject: [PATCH] fix num_cpu_blocks after changing block_size --- vllm/executor/openvino_executor.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/vllm/executor/openvino_executor.py b/vllm/executor/openvino_executor.py index d7caa772ee842..84059eacb7a4b 100644 --- a/vllm/executor/openvino_executor.py +++ b/vllm/executor/openvino_executor.py @@ -62,8 +62,9 @@ def __init__( if device_config.device.type == "cpu": if cache_config.block_size != 1: - print(f"Warning: CPU only support block_size = 1, current is {cache_config.block_size}, forced to 1.") + cache_config.num_cpu_blocks *= cache_config.block_size cache_config.block_size = 1 + print(f"Warning: CPU only support block_size = 1, it's forced to 1, num_cpu_blocks is set to {cache_config.num_cpu_blocks}.") self.block_size = cache_config.block_size self.num_gpu_blocks = cache_config.num_gpu_blocks self.num_cpu_blocks = cache_config.num_cpu_blocks