--output
Description
- For a model trained on an open-source framework:
This option specifies the directory (including the file name) of the generated offline model, for example, $HOME/module/out/tf_resnet50. The name of the generated model file is automatically suffixed with .om, for example, tf_resnet50.om or tf_resnet50_linux_x86_64.om. If the name of an .om file contains an OS and architecture, the .om file can be used only in an operating environment with such OS and architecture.
- For a single-operator description file (JSON format):
This option sets the directory of the generated single-operator model, for example, $HOME/singleop/out/op_model. The default naming rule of the generated model file is SN_opType_InputDescription(dataType_format_shape)_OutputDescription(dataType_format_shape). If you do not use this naming rule, you can specify the model file name by using the name attribute in the single-operator description file.
See Also
If the name of the .om model file generated by running the atc command contains an OS and architecture, but the OS and architecture are inconsistent with those in the model operating environment, you need to use this option in conjunction with --host_env_os and --host_env_cpu to set the OS type and architecture of the model operating environment.
Argument
Argument:
- Directory (including the file name) of an offline model converted from an open-source framework.
- Directory (including the file name) of a single-operator model converted from a single-operator description .json file.
Format: The directory (including the file name) can contain letters, digits, underscores (_), hyphens (-), periods (.), and Chinese characters.
Suggestions and Benefits
None
Example
- Caffe network:
--output=$HOME/module/out/caffe_resnet50
- TensorFlow network:
--output=$HOME/module/out/tf_resnet50
- ONNX network:
--output=$HOME/module/out/onnx_resnet50
- Single-operator description file:
--output=$HOME/singleop/out/op_model
Applicability
Dependencies and Restrictions
None