notes

Author: vpj

docs/diffusion/stable_diffusion/index.html

@@ -84,6 +84,7 @@ <h3><a href="latent_diffusion.html">Latent Diffusion Model</a></h3>
 <p>The core is the <a href="latent_diffusion.html">Latent Diffusion Model</a>. It consists of:</p>
 <ul><li><a href="model/autoencoder.html">AutoEncoder</a> </li>
 <li><a href="model/unet.html">U-Net</a> with <a href="model/unet_attention.html">attention</a></li></ul>
+<p>We have also (optionally) integrated <a href="https://github.com/HazyResearch/flash-attention">Flash Attention</a> into our <a href="model/unet_attention.html">U-Net attention</a> which lets you speed up the performance by close to 50% on an RTX A6000 GPU.</p>
 <p>The diffusion is conditioned based on <a href="model/clip_embedder.html">CLIP embeddings</a>.</p>
 <h3><a href="sampler/index.html">Sampling Algorithms</a></h3>
 <p>We have implemented the following <a href="sampler/index.html">sampling algorithms</a>:</p>