提交 3a8c622f 编写于 作者: W Wang Guibao 提交者: GitHub

Add documentation for runtime_thread_num (#8)

上级 eacebd34
......@@ -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
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册