Skip to content

Qcompiler/AccText2Img

Repository files navigation

An easy to use pluging for accelerate the SDXL. Just add the two line code before inference:

from mylinear import MixLinear_GEMM
torch.nn.Linear = MixLinear_GEMM

For FP16

srun -N 1 --gres=gpu:4090:1 python test.py

For mixed-int8

srun -N 1 --gres=gpu:4090:1 python test.py --mix_linear

About

Accelerate the text to image by mixed precsion computing

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors