search

castorini / daam

Diffusion attentive attribution maps for interpreting Stable Diffusion.
MIT License
689 stars 63 forks source link

RuntimeError: stack expects a non-empty TensorList #12

Closed federicotorrielli closed 1 year ago

federicotorrielli commented 1 year ago

image

When calling heat_map_lemma = tc.compute_global_heat_map(prompt)

daemon commented 1 year ago

Odd, can you share the prompt and surrounding code?

federicotorrielli commented 1 year ago

Sure, this is the pipe image And this is the call image

lemmas is just a list of string prompts. For the generator I used:

def initialize_generator(self):
    gen = torch.Generator(device='cuda')
    seed = 26111998
    return gen.manual_seed(seed)
daemon commented 1 year ago

Ah, I think I see the problem. DAAM doesn't support sliced attention. I'll release an update for that.

daemon commented 1 year ago

Fixed in latest release.