Failed to get convolution algorithm

[ttguy]

I had to add
tensorflow==2.6.0
imgaug<0.2.7,>=0.2.5
to end of the /content/faceswap/_requirements_base.txt
and make it load /content/faceswap/requirements_nvidia.txt
before I could run the training.

This seems to fix the following dependency issues I had:
tensorflow 2.7.0 requires keras<2.8,>=2.7.0rc0, but you have keras 2.6.0 which is incompatible.
tensorflow 2.7.0 requires tensorflow-estimator<2.8,~=2.7.0rc0, but you have tensorflow-estimator 2.6.0 which is incompatible.
albumentations 0.1.12 requires imgaug<0.2.7,>=0.2.5, but you have imgaug 0.2.9 which is incompatible.

But now training bombs with
tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found.

[[node villain/encoder/conv_128_0_conv2d/Conv2D_1 (defined at content/faceswap/plugins/train/trainer/_base.py:193) ]]
[[gradient_tape/LossWrapper_1/DSSIMObjective_2/Reshape_2/_30]]
(1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[node villain/encoder/conv_128_0_conv2d/Conv2D_1 (defined at content/faceswap/plugins/train/trainer/_base.py:193) ]]
0 successful operations.
0 derived errors ignored. [Op:__inference_train_function_14648]

Any hints on how to proceed to fixing this.