squall/flash-attention
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Link to Colossal-AI's stable diffusion in usage.md

Browse Source
This commit is contained in:
Tri Dao 2022-11-13 20:49:05 -08:00
parent 343492ec30
commit b92f2c3b67
1 changed files with 5 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
usage.md
View File

@ -59,6 +59,11 @@ yields the fastest BERT training on cloud instances in MLPerf training 2.0 (June
v0.7.0](https://github.com/huggingface/diffusers/releases/tag/v0.7.0).
Up to 2x faster inference and lower memory usage.
- Colossal-AI's
[implementation](https://github.com/hpcaitech/ColossalAI/tree/main/examples/images/diffusion)
of Stable Diffusion: with FlashAttention as one of its components, it speeds up
pretraining by up to 6.5x, and reduces the hardware cost of fine-tuning by 7x.
- Stable Diffusion inference from
[Labml.ai](https://twitter.com/labmlai/status/1573634095732490240): 50% speedup.