Skip to content

Commit ef7ddd0

Browse files
jonnorjonnor
committed
mnist_cnn: Enable output dequantize for quantized models
The outputs were frequently invalid without this, sometimes containing NaN values or huge numbers This brings the MNIST example from 60% to 94% accuracy
1 parent cc2cb6a commit ef7ddd0

File tree

5 files changed

+228
-227
lines changed

5 files changed

+228
-227
lines changed

0 commit comments

Comments
 (0)