Jovan Mitrevski
Jovan Mitrevski
I am including @AdrianAlan and @jmduarte because they were involved in the latest tracing updates.
I am not sure it's a complete fix. When I attempted the fairly straightforward fix of checking that it's not None, I ran into isuses when activation was a string.
We should validate the fix on this, as well.
Maybe it does fix it. I just checked out the PR, and I get no error. I thought that it complained about a string activation having `__name__` extracted, but I...
This has been fixed with the merge of #987 (which replaced #961)
> Do we need a corresponding fix for Quartus? > > https://github.com/fastmachinelearning/hls4ml/blob/f41951c57883022362dd4ffbe6ce3ed4efe36811/hls4ml/templates/quartus/firmware/nnet_utils/nnet_pooling.h#L38-L74 Actually I think it also needs to be done for vitis (which has its own implementation) in addition...
What did we decide to do about accum_t vs template matching?
My general feeling is to prefer the #973 style solution, though implementation details may still need to be addressed.
We decided to go with #973 so I will close this PR.
Still missing: tests, update documentation, and update other than keras parsing.