[doc] layergroup opt intro

Change-Id: I0797b73e4d020e9556da29d1c1a743b8c80a83ad
sophgo · Nov 22, 2024 · 537bd53 · 537bd53
1 parent 20e20b3
commit 537bd53
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/docs/developer_manual/source_en/03_user_interface.rst b/docs/developer_manual/source_en/03_user_interface.rst
@@ -595,7 +595,7 @@ Convert the mlir file into the corresponding model, the parameters are as follow
      - whether layer groups force group by cores, auto/true/false, default is auto
    * - opt
      - N
-     - Optimization type of LayerGroup, 1 or 2, default is 2
+     - Optimization type of LayerGroup, 1/2/3, default is 2. 1: Simple LayerGroup mode, all operators will be grouped as much as possible, and the compilation speed is faster; 2: Dynamic compilation calculates the global cycle optimal Group grouping, suitable for single-core inference graphs; 3: Linear programming LayerGroup mode, suitable for training graphs.
    * - addr_mode
      - N
      - set address assign mode ['auto', 'basic', 'io_alone', 'io_tag', 'io_tag_fuse'], if not set, auto as default

diff --git a/docs/developer_manual/source_zh/03_user_interface.rst b/docs/developer_manual/source_zh/03_user_interface.rst
@@ -608,7 +608,7 @@ model_deploy.py
      - layer groups是否根据core数目进行强制分组, 可选auto/true/false, 默认为auto
    * - opt
      - 否
-     - LayerGroup优化类型，可选1/2, 默认为2
+     - LayerGroup优化类型，可选1/2/3, 默认为2。1：简单LayerGroup模式，所有算子会尽可能做Group，编译速度较快；2：通过动态编译计算全局cycle最优的Group分组，适用于单核推理图；3：线性规划LayerGroup模式，适用于模型训练图编译。
    * - addr_mode
      - 否
      - 设置地址分配模式['auto', 'basic', 'io_alone', 'io_tag', 'io_tag_fuse'], 默认为auto