• State
  • Locked Locked
  • Replies 3 replies
  • Subscribers 100 subscribers
  • Views 148 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

AM69A: Relu6(Clip) issue in 32 vs 8 bit model inference

Liu Tsing-Yue
Intellectual 291 points
Part Number: AM69A

Hi, I'm using TIDL importer to import onnx models where

version = 09_00_00_02
SOC= AM69A
However, one of inference results seemed to have deviations between 32 bit and 8 bit model.
The 8-bit model output seemed  to Clipped the output value to 0~6, while the 32 bit model didn't Clipped the output value to 0~6.
What might be the reason here?
Here's the visualized model (svg files) of float and fixed models.
Thanks!