R

If tensorflow is unavailable because of system limitations then it should after 516271c1d5972fd6c73ccddd0f1326562eb740b1 hopefully not just give an error on start. However, as mentioned, only the wd* taggers work. If...

It does require the webui running somewhere, but it could be a docker container, vritualbox another server, e.g. on runpod, besides on your own computer. Via the fastapi it is...

ok, will check. I never touch that button. Thanks for reporting.

Removing the line will remove the attribute error, and I'll push that as a quick fix, but then the model wont unload. this exact line was one of the open...

Edit, actually it seems related to tensorflow only, so it's the models DeepDanbooruInterrogator and the experimental large_batch_interrogate. Any of the others should unload properly. Note that with my implementation you...

On [SO](https://stackoverflow.com/questions/52092999/how-to-unload-a-keras-tensorflow-model-from-memory) there is a solution mentioned: ```python from numba import cuda cuda.select_device(0) cuda.close() ``` But then the [numba documentation](http://numba.pydata.org/numba-doc/0.13/CUDADevice.html) is unclear how to reopen the same device again. And...

Thanks, this is why I placed it behind an experimental option in the settings. The Nvidia dependency for numba did occur to me, but at least for nvidia numba could...

Ah thanks again. I found [tf2onnx](https://github.com/onnx/tensorflow-onnx), which seems to do exactly this.

One option I am considering is implementing a Settings -> Tagger -> checkbox: convert tensorflow models to onnx. There might be checks necessary for updates, maybe keep the tensorflow next...

Interesting, I was implementing a similarity search within the batch of interrogated images that you have tagged.