T GATE Versions Save

T-GATE: Cross-Attention Makes Inference Cumbersome in Text-to-Image Diffusion Models

0.1.2

3 weeks ago

Bug Fixes:

fix issue #10 by considering the number of warm-up steps num_warmup_steps in the TgateLoader's.

0.1.1

1 month ago

All Commits

[pipeline_stable_diffusion_xl] fix issue #6 by adding a conditional check for denormalizing the latents.

0.1.0

1 month ago

These are the release notes of the TGATE library.

TGATE accelerates inferences of [PixArtAlphaPipeline], [StableDiffusionPipeline], and [StableDiffusionXLPipeline] by skipping the calculation of cross-attention once it converges. More details can be found at technical report.

🚀 Major Features

Training-Free.
Easily Integrate into Diffusers.
Only a few lines of code are required.
Complementary to DeepCache.
Friendly support Stable Diffusion pipelines, PixArt, and Latent Consistency Models.
10%-50% speed up for different models.

📖 Quick Start

Start by installing TGATE: pip install tgate
More examples can be found at here.