float 32 的 tflite 和 int 8 的tflite 轉換問題

kendryte / nncase

Open deep learning compiler stack for Kendryte AI accelerators ✨

Apache License 2.0

728 stars 181 forks source link

float 32 的 tflite 和 int 8 的tflite 轉換問題 #313

Closed KuoEuran closed 2 years ago

KuoEuran commented 3 years ago

Describe the bug 當我使用int 8的tflite進行kmodel轉換時,可以成功轉換,但無法在k210上運行

但當我使用float32的tflite進行kmodel轉換時,卻產生了error message: Fatal: inputs are not compatible to concat,無法進行轉換這令我很不解

Environment (please complete the following information):

OS: win10 nncase version： 0.2.0 beta4 DL Framework： tensorflow 2.5

再麻煩幫忙了,感謝各位

sunnycase commented 3 years ago

Please attach your origin tflite model. The kmodel doesn't have any graph info and can't be debugged.

zhangyang2057 commented 3 years ago

With the latest nncase, f32 model can be built successfully, but it failed with ptq mode.

zhangyang2057 commented 2 years ago

I use the latest nncase version(nncase v1.0.0 rc1), fp32 and int8 quantization can be built successfully. Please submit a new issue if you have any other problems.