Cannot run Codegen to generate code for other models

[csjimmylu]

I was trying to deploy a model with a different input shape to the STM32 board, but running this command raises NotImplementedError:

python examples/tiny_training.py -f full_bp-1x3x128x128-graph.json -D full_bp-1x3x128x128-params.pkl -QAS scale.json -m -g -d -FR

Where scale.json comes from img1 (highlighted) And both full_bp-1x3x128x128-graph.json & full_bp-1x3x128x128-params.pkl comes from img2 (highlighted)

These 3 files were generated accordingly from the tiny_training repo's compilation/readme.md

Any thoughts for this issue?

[meenchen]

Hi @729557989,

Thanks for reporting this issue. Could you also share the error message and stack trace for the issue?

[csjimmylu]

Does this help?

Commands I Ran:

1. conda activate tinyengine
2. export PYTHONPATH=${PYTHONPATH}:$(pwd)
3. python examples/tiny_training.py -f sm_assets/full_bp-1x3x128x128-graph.json -D sm_assets/full_bp-1x3x128x128-params.pkl -QAS sm_assets/scale.json -m -g -d -FR

Note: the 3 files accessed from 3. were all generated accordingly from the mit-han-lab/tiny_training repo's compilation/readme.md

The Error Message + Stack Trace:

[meenchen]

Hi @729557989,

We did not have the full opset to support bull_bp setting since the required memory would not fit in the SRAM. But feel free to try other sparse update schema and let us know if there are any issues.

[csjimmylu]

Hi @meenchen

Thanks for the reply! I have also tried sparse 49kbs ir (the minimal ram version) but this error still comes up.

[meenchen]

Hi @meenchen

Thanks for the reply! I have also tried sparse 49kbs ir (the minimal ram version) but this error still comes up.

Thanks for letting us know the issues. We will take a look into this.

[lyc18]

Hi  你好@meenchen Thanks for the reply! I have also tried sparse 49kbs ir (the minimal ram version) but this error still comes up.谢谢回复！我也尝试过稀疏 49kbs ir（最小 ram 版本），但仍然出现此错误。

Thanks for letting us know the issues. We will take a look into this.感谢您让我们了解问题。我们将对此进行研究。

Hi, is there a solution to this problem? I'm experiencing the same problem. I followed the README steps in tiny-training/compilation, but the problem also occurs when I try to convert the generated model in json format to a C++ version via examples/tiny_training.py.