Skip to content

Honor low_vram in ACE-Step load_model (fix OOM during latent caching) #911

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
SanDiegoDude wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Honor low_vram in ACE-Step load_model (fix OOM during latent caching) #911

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
13 changes: 11 additions & 2 deletions extensions_built_in/audio_models/ace_step/ace_step_15_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,8 +146,17 @@ def load_model(self):
self.vae = models["vae"]

# move back to device
self.model.to(device)
self.text_encoder.to(device)
if self.model_config.low_vram:
# Honor low_vram: keep the large transformer and text encoder on the CPU
# and let them be moved to the GPU on demand (see get_prompt_embeds,
# get_noise_prediction, generate_single_audio). Only the VAE needs to be
# resident for latent caching, which otherwise OOMs when the full quantized
# DiT + text encoder are also occupying VRAM.
self.model.to("cpu")
self.text_encoder.to("cpu")
else:
self.model.to(device)
self.text_encoder.to(device)
self.vae.to(device)
self.tokenizer = models["tokenizer"]

Expand Down