opt tool - Only version 0 is supported

[Qengineering]

Having issue #4560, I followed the given answer.

Hi, you need to convert the original model again to make it workable on Paddle-Lite v2.7.
You can follow steps below:

1、 Clone Paddle-Lite and checkout to v2.7-beta: cd Paddle-Lite && git checkout v2.7-beta
2、Compile opt from v2.7-beta: ./lite/tools/build.sh build_optimize_tool
3、 Convert original model by opt of v2.7: ./opt --model_dir=your_model_path --valid_targets=your_target -->optimize_out=opt_model_v2.7
4、 Compile inference library from v2.7
5、Run opt_model_v2.7 on inference library from v2.7

1 + 2) Done -> succes.
3) Get original model

Install PaddlePaddle v2.0.0-rc0 + PaddleHub v1.8.1 ->succes.

$ python3
>>> import paddlehub as hub
>>> import paddle
>>> paddle.enable_static()
>>> pyramidbox_lite_mobile_mask = hub.Module(name="pyramidbox_lite_mobile_mask")
>>> pyramidbox_lite_mobile_mask.save_inference_model(dirname="test_program")

Got test_program

.
├── mask_detector
│   ├──__model__
│   └──__params__
└── pyramidbox_lite
    ├── __model__
    └──__params__

Loaded the mask_detector in opt tool and got this error:

Note: same Only version 0 is supported error is given with the model is generated by PaddlePaddle v1.8.5, as being the latest stable version.

[Qengineering]

Found a workaround.
Instead of --model_dir, used --model_file= and --param_file=.

[Qengineering]

However, the generated Paddle-Lite model can't be run due to no Op found for feed error.

[DannyIsFunny]

Are you linking your project to Paddle-Lite static library libpaddle_api_light_bundled.a ？In this case, you need to include extra head files as listed below:

/////////////////////////////////////////////////////////////////////////
// If this demo is linked to static libaray:libpaddle_api_light_bundled.a
// , you should include `paddle_use_ops.h` and `paddle_use_kernels.h`.
/////////////////////////////////////////////////////////////////////////
#include "paddle_use_kernels.h"  // NOLINT
#include "paddle_use_ops.h"      // NOLINT

[Qengineering]

Works like a charm!
Thanks @DannyIsFunny !