Closed skywo1f closed 3 years ago
should be fixed with https://github.com/deepglugs/dalle/commit/36ee7eba025c11e0d243d7fd369cec44727210e8
seems to have fixed the problem, though now I am getting a new problem:
~/neural-nets/dalle$ ./train.sh
loading state dict from vae.pt
preloading txt files...
processing 72496/72496
done preloading txt onehots
could not find 94839-004-61CB5BCA in preloaded txts
Traceback (most recent call last):
File "dalle.py", line 617, in
this is usually caused by having image files without a corresponding text file. For example there is a 94839-004-61CB5BCA.png without a 94839-004-61CB5BCA.txt.
It says it right there haha. I should probably learn to read.
I was able to train the vae, but I am now stuck on the dalle training step. It is telling me: loading state dict from vae.pt preloading txt files... processing 72496/72496 done preloading txt onehots Train Epoch: 1 [0/72496 (0%)] Loss: 0.569296 Traceback (most recent call last): File "dalle.py", line 615, in main() File "dalle.py", line 608, in main train_dalle(vae, args) File "dalle.py", line 331, in train_dalle if clip is not None: UnboundLocalError: local variable 'clip' referenced before assignment