I was discussing model downloads privately with another beta tester, and he noted that a number of the files that I suggested did not exist. I had not tried to download any files other than the onnx16 versions, since that’s what my systems call for. He tried some of the others and these are reported errors:
tensorflow models do not appear to exist. This is not a big deal because the .json files indicate only 288x288 for those, so they are presumably distributed. I also note, in looking at the json for tensorflow the file names aren’t consistent between models. In some they are fp32-288x288 and in others they are 288x288-fp32.
Due to its very resource heavy nature, lack of performance, newer hardware support. We have moved away from Tensorflow. The Tensorflow entries in the json file are merely placeholder used for internal testing and benchmarking. So the tf models will not be available for downloads and won’t run with VEAI 1.7 and beyond.
Thanks for letting us know that some of the ahq files were not publicly accessible. The issue has been fixed. If you find anything else missing let us know.
Does Topaz care if I share my method for downloading all files? I am not surprised that someone has asked in the public forum, but I didn’t want to post the method publicly without asking first.
Yes, you can post your method publicly. Hopefully, this time it is fixed for good, fp32 is only needed for old AMD and Nvidia cards. Old AMD cards have a 100% decrease in speed with FP16 and old Nvidia GTX GPUs have 5-10%.