Skip to content

fix: move conditioning tensors to model device when partial loading enabled #8970

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
aayushbaluni wants to merge 1 commit into
base: main
Choose a base branch
from
+14 −5
Open

fix: move conditioning tensors to model device when partial loading enabled #8970

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 14 additions & 5 deletions invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -337,9 +337,13 @@ def _apply_standard_conditioning(
)
cross_attention_kwargs["percent_through"] = step_index / total_step_count

uncond_embeds = conditioning_data.uncond_text.embeds.to(x.device)
cond_embeds = conditioning_data.cond_text.embeds.to(x.device)
both_conditionings, encoder_attention_mask = self._concat_conditionings_for_batch(
conditioning_data.uncond_text.embeds, conditioning_data.cond_text.embeds
uncond_embeds, cond_embeds
)
if added_cond_kwargs is not None:
added_cond_kwargs = {k: v.to(x.device) for k, v in added_cond_kwargs.items()}
both_results = self.model_forward_callback(
x_twice,
sigma_twice,
Expand Down Expand Up @@ -428,11 +432,15 @@ def _apply_standard_conditioning_sequentially(
)
cross_attention_kwargs["percent_through"] = step_index / total_step_count

uncond_embeds = conditioning_data.uncond_text.embeds.to(x.device)
if added_cond_kwargs is not None:
added_cond_kwargs = {k: v.to(x.device) for k, v in added_cond_kwargs.items()}

# Run unconditioned UNet denoising (i.e. negative prompt).
unconditioned_next_x = self.model_forward_callback(
x,
sigma,
conditioning_data.uncond_text.embeds,
uncond_embeds,
cross_attention_kwargs=cross_attention_kwargs,
down_block_additional_residuals=uncond_down_block,
mid_block_additional_residual=uncond_mid_block,
Expand Down Expand Up @@ -465,8 +473,8 @@ def _apply_standard_conditioning_sequentially(
added_cond_kwargs = None
if conditioning_data.is_sdxl():
added_cond_kwargs = {
"text_embeds": conditioning_data.cond_text.pooled_embeds,
"time_ids": conditioning_data.cond_text.add_time_ids,
"text_embeds": conditioning_data.cond_text.pooled_embeds.to(x.device),
"time_ids": conditioning_data.cond_text.add_time_ids.to(x.device),
}

# Prepare prompt regions for the conditioned pass.
Expand All @@ -476,11 +484,12 @@ def _apply_standard_conditioning_sequentially(
)
cross_attention_kwargs["percent_through"] = step_index / total_step_count

cond_embeds = conditioning_data.cond_text.embeds.to(x.device)
# Run conditioned UNet denoising (i.e. positive prompt).
conditioned_next_x = self.model_forward_callback(
x,
sigma,
conditioning_data.cond_text.embeds,
cond_embeds,
cross_attention_kwargs=cross_attention_kwargs,
down_block_additional_residuals=cond_down_block,
mid_block_additional_residual=cond_mid_block,
Expand Down
Loading