diff --git a/doc/SERVING_CONFIGURE.md b/doc/SERVING_CONFIGURE.md index f5887f5cd139b1e70b49f4eee2e2552658692103..f34436a064963ae25411af53b3116f00b9a493de 100644 --- a/doc/SERVING_CONFIGURE.md +++ b/doc/SERVING_CONFIGURE.md @@ -165,7 +165,7 @@ reloadable_type: 检查reload条件:timestamp_ne/timestamp_gt/md5sum/revision/ model_data_path: 模型文件路径 -runtime_thread_num: 若大于0, 则启用bsf多线程调度框架,在每个预测bthread worker内启动多线程预测。 +runtime_thread_num: 若大于0, 则启用bsf多线程调度框架,在每个预测bthread worker内启动多线程预测。要注意的是,当启用worker内多线程预测,workflow中OP需要用Serving框架的BatchTensor类做预测的输入和输出 (predictor/framework/infer_data.h, `class BatchTensor`)。 batch_infer_size: 启用bsf多线程预测时,每个预测线程的batch size