fix bugs related to variable prompt lengths

This commit is contained in:
AUTOMATIC 2022-10-08 15:25:59 +03:00
parent 4999eb2ef9
commit 77f4237d1c
2 changed files with 37 additions and 12 deletions

View File

@ -89,7 +89,6 @@ class StableDiffusionModelHijack:
layer.padding_mode = 'circular' if enable else 'zeros'
def tokenize(self, text):
max_length = opts.max_prompt_tokens - 2
_, remade_batch_tokens, _, _, _, token_count = self.clip.process_text([text])
return remade_batch_tokens[0], token_count, get_target_prompt_token_count(token_count)
@ -174,7 +173,8 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
if line in cache:
remade_tokens, fixes, multipliers = cache[line]
else:
remade_tokens, fixes, multipliers, token_count = self.tokenize_line(line, used_custom_terms, hijack_comments)
remade_tokens, fixes, multipliers, current_token_count = self.tokenize_line(line, used_custom_terms, hijack_comments)
token_count = max(current_token_count, token_count)
cache[line] = (remade_tokens, fixes, multipliers)
@ -265,15 +265,19 @@ class FrozenCLIPEmbedderWithCustomWords(torch.nn.Module):
if len(used_custom_terms) > 0:
self.hijack.comments.append("Used embeddings: " + ", ".join([f'{word} [{checksum}]' for word, checksum in used_custom_terms]))
position_ids_array = [min(x, 75) for x in range(len(remade_batch_tokens[0])-1)] + [76]
target_token_count = get_target_prompt_token_count(token_count) + 2
position_ids_array = [min(x, 75) for x in range(target_token_count-1)] + [76]
position_ids = torch.asarray(position_ids_array, device=devices.device).expand((1, -1))
tokens = torch.asarray(remade_batch_tokens).to(device)
remade_batch_tokens_of_same_length = [x + [self.wrapped.tokenizer.eos_token_id] * (target_token_count - len(x)) for x in remade_batch_tokens]
tokens = torch.asarray(remade_batch_tokens_of_same_length).to(device)
outputs = self.wrapped.transformer(input_ids=tokens, position_ids=position_ids)
z = outputs.last_hidden_state
# restoring original mean is likely not correct, but it seems to work well to prevent artifacts that happen otherwise
batch_multipliers = torch.asarray(batch_multipliers).to(device)
batch_multipliers_of_same_length = [x + [1.0] * (target_token_count - len(x)) for x in batch_multipliers]
batch_multipliers = torch.asarray(batch_multipliers_of_same_length).to(device)
original_mean = z.mean()
z *= batch_multipliers.reshape(batch_multipliers.shape + (1,)).expand(z.shape)
new_mean = z.mean()

View File

@ -142,6 +142,16 @@ class VanillaStableDiffusionSampler:
assert all([len(conds) == 1 for conds in conds_list]), 'composition via AND is not supported for DDIM/PLMS samplers'
cond = tensor
# for DDIM, shapes must match, we can't just process cond and uncond independently;
# filling unconditional_conditioning with repeats of the last vector to match length is
# not 100% correct but should work well enough
if unconditional_conditioning.shape[1] < cond.shape[1]:
last_vector = unconditional_conditioning[:, -1:]
last_vector_repeated = last_vector.repeat([1, cond.shape[1] - unconditional_conditioning.shape[1], 1])
unconditional_conditioning = torch.hstack([unconditional_conditioning, last_vector_repeated])
elif unconditional_conditioning.shape[1] > cond.shape[1]:
unconditional_conditioning = unconditional_conditioning[:, :cond.shape[1]]
if self.mask is not None:
img_orig = self.sampler.model.q_sample(self.init_latent, ts)
x_dec = img_orig * self.mask + self.nmask * x_dec
@ -221,18 +231,29 @@ class CFGDenoiser(torch.nn.Module):
x_in = torch.cat([torch.stack([x[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [x])
sigma_in = torch.cat([torch.stack([sigma[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [sigma])
cond_in = torch.cat([tensor, uncond])
if shared.batch_cond_uncond:
x_out = self.inner_model(x_in, sigma_in, cond=cond_in)
if tensor.shape[1] == uncond.shape[1]:
cond_in = torch.cat([tensor, uncond])
if shared.batch_cond_uncond:
x_out = self.inner_model(x_in, sigma_in, cond=cond_in)
else:
x_out = torch.zeros_like(x_in)
for batch_offset in range(0, x_out.shape[0], batch_size):
a = batch_offset
b = a + batch_size
x_out[a:b] = self.inner_model(x_in[a:b], sigma_in[a:b], cond=cond_in[a:b])
else:
x_out = torch.zeros_like(x_in)
for batch_offset in range(0, x_out.shape[0], batch_size):
batch_size = batch_size*2 if shared.batch_cond_uncond else batch_size
for batch_offset in range(0, tensor.shape[0], batch_size):
a = batch_offset
b = a + batch_size
x_out[a:b] = self.inner_model(x_in[a:b], sigma_in[a:b], cond=cond_in[a:b])
b = min(a + batch_size, tensor.shape[0])
x_out[a:b] = self.inner_model(x_in[a:b], sigma_in[a:b], cond=tensor[a:b])
denoised_uncond = x_out[-batch_size:]
x_out[-uncond.shape[0]:] = self.inner_model(x_in[-uncond.shape[0]:], sigma_in[-uncond.shape[0]:], cond=uncond)
denoised_uncond = x_out[-uncond.shape[0]:]
denoised = torch.clone(denoised_uncond)
for i, conds in enumerate(conds_list):