VeriSilicon / acuitylite

Acuitylite is an end-to-end neural network deployment tool
https://verisilicon.github.io/acuitylite
MIT License
16 stars 5 forks source link

Usefull or just for fun! #9

Closed 2050airobert closed 1 month ago

2050airobert commented 2 years ago

hi, 1 I don not know what you said below. "Tips: You can export a TFLite app and using tflite-vx-delegate to run on 2 TIM-VX if the exported TIM-VX app does not meet your requirements." 2 There is a realworld quesion, if I have a more layers and newest tflite model, could I transfer it by this acuitylite tool to generate the testcase of my tflite mode? Or it supports only few and old cnn models? 3 Is there any detailed step to intruduce the use of this tool? BR

2050airobert commented 2 years ago

hi,any one could help ? tks

lileiigithub commented 2 years ago

Thanks for your feedback.

  1. Acuitylite can export the timvx case directly for TIM-VX, but when the exporter does something wrong, you can also try exporting tflite model for running on TIM-VX by using tflite-vx-delegate.
  2. Acuitylite dependence on tensorflow2.8 in the current version, so it will export the tflite model version of tensorflow2.8.
  3. You can refer to the demo and document on https://verisilicon.github.io/acuitylite/.
2050airobert commented 2 years ago

tks,you mean the tflite-vx-delegate will generate the demo code of my newest cnn model just like what NBG project could do ? After transfering with tflite-vx-delegate,I can run the demo of my code including pre and post processing of my model ?

lileiigithub commented 2 years ago

I think yes, but you can refer to https://github.com/VeriSilicon/tflite-vx-delegate for more info. And you can make an issue for more questions under tflite-vx-delegate repo.

2050airobert commented 2 years ago

I did! Yet no one answers me such as your kind hearted reply! https://github.com/VeriSilicon/tflite-vx-delegate/issues/92

lileiigithub commented 2 years ago

I am sorry for that. And I will ask for help from others tomorrow workday because I don't know much about tflite-vx-delegate project. BR

2050airobert commented 2 years ago

I see already. Thankful to you , you are so nice and hardworking guy! Looking forward to you good feedback !

2050airobert commented 1 year ago

hi, Why the number of users of acuitylite tool is so small? No issues or questions at all! I tried some new cnn models converting to int8 model on the a311d NPU with this tools, yet much error occurs still now! Could we assure the tool is situable for nearly the most of new CNN model all the world? If I have a selfdefined yet still format onnx model, could I get the int8 model easily and no error at all? Right ? BR

geneyoung commented 1 year ago

hi, Why the number of users of acuitylite tool is so small? No issues or questions at all! I tried some new cnn models converting to int8 model on the a311d NPU with this tools, yet much error occurs still now! Could we assure the tool is situable for nearly the most of new CNN model all the world? If I have a selfdefined yet still format onnx model, could I get the int8 model easily and no error at all? Right ? BR

Hi, Sorry for you any inconvinence. Generally, our tool can support many popular ONNX models, we don't know what concrete problems you met. Do you crashed at load/quantize/export stage? could you please paste you error log for us to debug? Thanks in advance for your feedback.