search

espressif / esp-dl

Espressif deep-learning library for AIoT applications
MIT License
549 stars 118 forks source link

TVM generated code is not correct (output_data.h) (AIV-635) #128

Open Farzinkh opened 1 year ago

Farzinkh commented 1 year ago

Hi, thank you for adding this TVM feature. However, I think the generated code is not what it is supposed to be. Can you help me to tackle this problem? output_data.h content:
#include <tvmgen_default.h> const size_t output_len = 10; const static __attribute__((aligned(16))) float output_data[10] ={}; The ONNX model I am trying is for the classification of the MNIST dataset with only 2 Dense layers. And I found another problem in export_onnx_model.py changing line 12 to: esp_dl_library_path = os.environ["ESP_DL_PATH"]+'/' This will solve the error of not finding files when calling this script in another one.

Auroragan commented 1 year ago

Hi, Could you give more info about: 'the generated code is not what it is supposed to be'? What is your expected result?

And thanks for the advice.

Farzinkh commented 1 year ago

Hi, I don't even know How in the first place I was capable of running export_onnx_model currently I am receiving this error esp-dl/tools/tvm/python/tvm/relay/ty.py", line 26, in <module> Any = _ffi_api.Any AttributeError: module 'tvm.relay._ffi_api' has no attribute 'Any'