brevitas
brevitas copied to clipboard
loss nan
Hi, I finished a network with weight_bit_width=2, but when I set weight_bit_width=1, this warning come up.
What is wrong?
And how can I fix it?
Looking forward to your reply!
Thanks
maybe the weight width is too small
Hi,thanks for your reply! I do think so but I want to train a net with 1 bit width, what can I do to achieve this goal? Is there any solution? Looking forward to your reply!
I have not explored this part yet, so I can not help
Thanks for your reply! I was troubled by this. If there are any questions I think we can discuss about it.
sure, but I guess discussions would be a better option than issues
I have run into this issue even with 8-bit weights or 16-bit weights. See my model at https://github.com/Xilinx/brevitas/discussions/679, but many variations of it have the same behavior.