Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' into lstein/model-manager-route-enhancements

Browse Source
This commit is contained in:
blessedcoolant
2023-07-15 16:56:54 +12:00
committed by GitHub
parent 2faa7cee37 194434dbfa
commit 808b2de709
14 changed files with 327 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py
View File

@ -241,11 +241,45 @@ class InvokeAIDiffuserComponent:
def _apply_standard_conditioning(self, x, sigma, unconditioning, conditioning, **kwargs):
# fast batched path
def _pad_conditioning(cond, target_len, encoder_attention_mask):
conditioning_attention_mask = torch.ones((cond.shape[0], cond.shape[1]), device=cond.device, dtype=cond.dtype)
if cond.shape[1] < max_len:
conditioning_attention_mask = torch.cat([
conditioning_attention_mask,
torch.zeros((cond.shape[0], max_len - cond.shape[1]), device=cond.device, dtype=cond.dtype),
], dim=1)
cond = torch.cat([
cond,
torch.zeros((cond.shape[0], max_len - cond.shape[1], cond.shape[2]), device=cond.device, dtype=cond.dtype),
], dim=1)
if encoder_attention_mask is None:
encoder_attention_mask = conditioning_attention_mask
else:
encoder_attention_mask = torch.cat([
encoder_attention_mask,
conditioning_attention_mask,
])
return cond, encoder_attention_mask
x_twice = torch.cat([x] * 2)
sigma_twice = torch.cat([sigma] * 2)
encoder_attention_mask = None
if unconditioning.shape[1] != conditioning.shape[1]:
max_len = max(unconditioning.shape[1], conditioning.shape[1])
unconditioning, encoder_attention_mask = _pad_conditioning(unconditioning, max_len, encoder_attention_mask)
conditioning, encoder_attention_mask = _pad_conditioning(conditioning, max_len, encoder_attention_mask)
both_conditionings = torch.cat([unconditioning, conditioning])
both_results = self.model_forward_callback(
x_twice, sigma_twice, both_conditionings, **kwargs,
x_twice, sigma_twice, both_conditionings,
encoder_attention_mask=encoder_attention_mask,
**kwargs,
)
unconditioned_next_x, conditioned_next_x = both_results.chunk(2)
return unconditioned_next_x, conditioned_next_x