bhack
bhack
> you are trying to cast a list with tensor in it, into a tensor. You should just need to unpack it correctly. Yes I found that one but it...
I have at least a running draft. Let me know if you want to review with a PR in this repo.
I've tried to upscale the decoder more smoothly with some extra layers (512 and 1024) up to `1024x1024` instead of the original `256x256` + pure interpolation and I have seen...
> Maybe a few regular convolutions in between the upscaling steps could help blend things better spatially It is what I have tried to not invalidate the pretrained checkpoint part...
Is this going in the 2.7.0 release?
`class BahdanauAttention(_BaseAttentionMechanism):` And in`_BaseAttentionMechanism:` > Also note that this layer does not work with Keras model when `model.compile(run_eagerly=True)` due to the fact that this layer is stateful. The support for...
I don't know if in your case could be resolved with manual memory reset in the PR introduced fixing https://github.com/tensorflow/addons/issues/535
Are you planning to release wheels after merge?
I have the same issue with today pytorch nightly and ao nightly. ```python TypeError: _make_wrapper_subclass(): argument 'dtype' must be torch.dtype, not torch._C._TensorMeta from user code: File "/opt/conda/lib/python3.11/site-packages/torchao/quantization/autoquant.py", line 945, in...
Any news on this?