Skip to content

ltx23: compile the VAE in the SPEED profile (7.3s -> ~5.0s) #18

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

Merged
Johan-de-R merged 1 commit into from
Jul 9, 2026
Merged

ltx23: compile the VAE in the SPEED profile (7.3s -> ~5.0s) #18

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
10 changes: 9 additions & 1 deletion examples/diffusers/ltx23/streaming_speed.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,12 +50,20 @@ generator:
pin_cpu_memory: true

# internal: gpu_pool.py:251-258 + the just-uncommented "mode" line
# — torch.compile on for transformer + text encoder, inductor
# — torch.compile on for transformer + text encoder + VAE, inductor
# backend, fullgraph for graph-break debugging, max-autotune-
# no-cudagraphs for the kernel sweep, dynamic off (static shapes).
# vae_enabled matches FastVideo's batch path (video_generation.py sets
# vae_enabled=ENABLE_TORCH_COMPILE, default on); measured A/B on the FA4
# image (2026-07-09, di-slc-140 B200) cut warm gen 7.16s -> 5.02s (~2.1s,
# ~30%) by compiling the VAE decode. Quality-neutral (torch.compile is
# numerically ~identical). Requires a re-bake so the Mega-Cache captures
# the VAE graphs (save_cache_artifacts snapshots them after the first
# forward); a stale blob just cold-compiles the VAE, never wedges.
compile:
enabled: true
text_encoder_enabled: true
vae_enabled: true
backend: inductor
fullgraph: true
mode: max-autotune-no-cudagraphs
Expand Down
Loading