# 3.2.6 编译ONNX模型

BMNETO 是针对 ONNX 的模型编译器，可以把 ONNX 格式的 model 经过图编译优化后，转换成 BMRuntime 所需的文件。在编译模型的同时，可选择将每一个操作的 NPU 模型计算结果和 CPU 的计算结果进行对比， 保证正确性。

* 命令行形式：

```bash
python3 -m bmneto [--model=<path>] \ 
                  [--input_names=<string>] \ 
                  [--shapes=<string>] \ 
                  [--outdir=<path>] \ 
                  [--target=<name>] \ 
                  [--net_name=<name>] \ 
                  [--opt=<value>] \ 
                  [--dyn=<bool>] \ 
                  [--cmp=<bool>] \ 
                  [--mode=<string>] \ 
                  [--descs=<string>] \ 
                  [--enable_profile=<bool>] \ 
                  [--output_names=<string>] \ 
                  [--list_ops]
```

参数介绍：

<table data-header-hidden><thead><tr><th width="205.66666666666666">args</th><th>type</th><th>Description</th></tr></thead><tbody><tr><td>args</td><td>type</td><td>Description</td></tr><tr><td>model</td><td>string</td><td><strong>Necessary.</strong>ONNX model (.onnx) path</td></tr><tr><td>input_names</td><td>string</td><td><strong>Optional.</strong>Set name of all network inputs one by one in sequence. Format “name1,name2,name3”</td></tr><tr><td>shapes</td><td>string</td><td><strong>Necessary.</strong> Shapes of all inputs, default use the shape in prototxt, format [[x,x,x,x],[x,x],…], these correspond to inputs one by one in sequence</td></tr><tr><td>outdir</td><td>string</td><td><strong>Necessary.</strong> Output directory</td></tr><tr><td>target</td><td>string</td><td><strong>Necessary.</strong> Option: BM1682, BM1684; default: BM1682</td></tr><tr><td>net_name</td><td>string</td><td><strong>Optional.</strong> Name of the network, default use the name in onnx path</td></tr><tr><td>opt</td><td>int</td><td><strong>Optional.</strong> Optimization level. Option: 0, 1, 2, default 1.</td></tr><tr><td>dyn</td><td>bool</td><td><strong>Optional.</strong> Use dynamic compilation, default false.</td></tr><tr><td>cmp</td><td>bool</td><td><strong>Optional.</strong>Check result during compilation. Default: true</td></tr><tr><td>mode</td><td>string</td><td><strong>Optional.</strong> Set bmnetc mode. Option: compile, GenUmodel. Default: compile.</td></tr><tr><td>descs</td><td>string</td><td><strong>Optional.</strong> Describe data type and value range of some input in format: "[ index, data format, lower bound, upper bound ]", where data format could be fp32, int64. For example, "[0, int64, 0, 100]", meaning input of index 0 has data type as int64 and values in [0, 100). If no description of some input given, the data type will be fp32 as default and uniformly distributed in 0 ~ 1.</td></tr><tr><td>enable_profile</td><td>bool</td><td><strong>Optional.</strong> Enable profile log. Default: false</td></tr><tr><td>output_names</td><td>string</td><td><strong>Optional.</strong> Set name of all network outputs one by one in sequence. Format "name1,name2,name2"</td></tr><tr><td>list_ops</td><td></td><td><strong>Optional.</strong> List supported ops.</td></tr></tbody></table>

* Python模式：

```python
import bmneto
## compile fp32 model
bmneto.compile(
    model = "/path/to/.pth", ## Necessary
    outdir = "xxx", ## Necessary
    target = "BM1684", ## Necessary
    shapes = [[x,x,x,x], [x,x,x]], ## Necessary
    net_name = "name", ## Necessary
    input_names = ['name0','name1'] ## Necessary
    opt = 2, ## optional, if not set, default equal to 1
    dyn = False, ## optional, if not set, default equal to False
    cmp = True, ## optional, if not set, default equal to True
    enable_profile = True ## optional, if not set, default equal to False
    descs = [[0, int, 0, 128]] ## optional, if not set, default equal to [[x, float, 0,
,→ 1]]
    output_names = ['oname0','oname1'] ## optional, if not set, default equal to graph␣ ,→output names
)
```

bmneto若成功，输出的 log 最后会看到以下信息：

```bash
######################################
# Store bmodel of BMCompiler.
######################################
```

bmneto成功后，将在指定的文件夹中生成一个compilation.bmodel的文件，该文件则是转换成功的 bmodel，用户可以重命名。