DreamBooth runs with runwayml/stable-diffusion-v1-5 but produces an empty list

[atonalfreerider]

Switching to runwayml/stable-diffusion-v1-5 from stabilityai/stable-diffusion-2-1-base is the only way that I can get DreamBooth to run on my 11GB VRAM GPU, but it produces no results

  File "/home/john/Desktop/3DPose/PuzzleAvatar/multi_concepts/train.py", line 1165, in aggregate_attention
    out = torch.cat(out, dim=0)
RuntimeError: torch.cat(): expected a non-empty list of Tensors

if I skip DreamBooth altogether, I can get the geometry and texture outputs on my custom data, but even after 20k steps of training, the model looks terrible.

Is DreamBooth necessary for a good output?